diff --git a/optimized/R2V/README.md b/optimized/R2V/README.md index 8127530..288c161 100644 --- a/optimized/R2V/README.md +++ b/optimized/R2V/README.md @@ -6,6 +6,7 @@ MiniMax baseline is a validated 10-second, 960×544 workflow; the earlier 3-second/480p efficiency tests remain recorded as historical comparisons. - [`minimax/`](minimax/) — MiniMax H3 R2V base and efficiency matrix +- [`minimax/sparse-latent-upscale/`](minimax/sparse-latent-upscale/) — supplied Fast H3 sparse-attention and tiled latent-upscale study Current baseline: MiniMax MM-R2V-68, Turbo LoRA, SLA 0.85, Comfy Kitchen attention, Low-VRAM Attention, and Chunk FeedForward at 8 steps, Res diff --git a/optimized/R2V/minimax/README.md b/optimized/R2V/minimax/README.md index d2bdbf1..00842cf 100644 --- a/optimized/R2V/minimax/README.md +++ b/optimized/R2V/minimax/README.md @@ -35,6 +35,13 @@ substitute another configuration. - `variables.json` — fixed comparison variables - `results/` — labelled outputs and matching JSON metadata +## New sparse-attention latent-upscale variant + +The supplied Fast H3 sparse-attention and tiled latent-upscale study is +documented separately in [`sparse-latent-upscale/`](sparse-latent-upscale/). +It is intentionally not promoted over the current Turbo baseline until the +controlled upscale, tile and initial-dimension tests are complete. + ## Current best **MM-R2V-58** remains the historical 3-second baseline: Turbo LoRA enabled, 8 diff --git a/optimized/R2V/minimax/sparse-latent-upscale/README.md b/optimized/R2V/minimax/sparse-latent-upscale/README.md new file mode 100644 index 0000000..4a5c62b --- /dev/null +++ b/optimized/R2V/minimax/sparse-latent-upscale/README.md @@ -0,0 +1,72 @@ +# MiniMax H3 sparse-attention latent-upscale study + +This workflow variant is based on the supplied CSGlide H3 workflow. It uses: + +- `fastvideo_fasth3_8step_v2_pruned_int8_convrot.safetensors` +- Block Sparse Attention with VSA selection +- Comfy Kitchen attention, MiniMax low-VRAM attention and chunked feed-forward +- 8-step Euler + beta initial generation +- MiniMax latent upscaling with tiled spatial splitting and latent refinement +- The supplied VAE, audio VAE, references, seed and one approximately 10-second clip + +The workflow is preserved in `workflow-source.json`; `workflow.json` is the tracked test copy with the enhanced common prompt and corrected reference filenames. + +## Fixed test input + +- **References:** `references/Scene022.png` as `@image1`, `references/Char034_TurnaroundFull_001.png` as `@image2` +- **Clip:** 243 frames at 24 fps, approximately 10.125 seconds +- **Seed:** `317182692766673` for the initial sampler; keep it unchanged in every test +- **Initial sampler:** Euler, beta scheduler, 8 steps, denoise 1.0 +- **Latent refinement:** SA solver, simple scheduler, 3 steps, denoise 0.2, CFG 1 +- **Prompt:** `prompt.txt`; do not change it between tests + +## Test matrix + +Every test changes only the listed variables. Record runtime, peak VRAM, success/failure, output dimensions, visible seams, detail retention, anatomy, motion stability and audio output in the matching result JSON. + +### Phase A — Upscale ceiling + +| ID | Initial latent | Target latent | Tile | Purpose | +|---|---:|---:|---:|---| +| SLU-01 | 672×384 | 1344×768 | supplied baseline tile | Reproduce the supplied good result | +| SLU-02 | 672×384 | 1920×1080 | supplied baseline tile | Test the desired 1080p endpoint | +| SLU-03 | 672×384 | 1680×960 | supplied baseline tile | Intermediate fallback if 1080p fails or is too slow | +| SLU-04 | 672×384 | 1536×864 | supplied baseline tile | Practical quality/runtime fallback | + +### Phase B — Tile size + +Use the best successful target from Phase A and compare tile sizes while keeping overlap and all other settings fixed: + +- SLU-05: 512×512 tiles +- SLU-06: 640×640 tiles +- SLU-07: 768×768 tiles +- SLU-08: 864×480-style wide tiles from the supplied workflow +- SLU-09: full-frame/no split if VRAM permits + +For each tile, preserve 128-pixel spatial overlap where possible, then check seam visibility, runtime and peak VRAM. + +### Phase C — Initial generation dimensions + +Use the best target and tile combination: + +- SLU-10: 672×384 supplied baseline +- SLU-11: 768×432 +- SLU-12: 832×480 supplied alternate resolution node +- SLU-13: 960×544 +- SLU-14: 1024×576 if the 8GB card remains stable + +The initial aspect ratio must remain 16:9 and the same final target must be used for fair comparison. + +### Phase D — Combined candidates + +- SLU-15: best upscale ceiling + best tile + best initial dimensions +- SLU-16: quality-first combination, accepting a longer runtime if the visual improvement is clear +- SLU-17: practical production combination targeting a predictable sub-hour render + +## First two runs + +Run **SLU-01** and **SLU-02** first. SLU-01 validates the supplied configuration. SLU-02 directly tests whether 1080p is feasible on the RTX 3070. If SLU-02 exhausts VRAM, stalls or becomes impractically slow, stop that run and continue with SLU-03/04 rather than changing the prompt or seed. + +## Evidence and verdict + +The test is **VALIDATED** only when a configuration completes with a usable video and recorded measurements. A completed render with unacceptable seams, detail loss or motion instability is **PARTIAL**, not a success. Keep all result JSON and MP4 files under `results/`. diff --git a/optimized/R2V/minimax/sparse-latent-upscale/prompt.txt b/optimized/R2V/minimax/sparse-latent-upscale/prompt.txt new file mode 100644 index 0000000..62df447 --- /dev/null +++ b/optimized/R2V/minimax/sparse-latent-upscale/prompt.txt @@ -0,0 +1,18 @@ +detailed_description: +@image1 is the environment and master scene reference. Preserve the wheat field, distant tree line, daylight, atmospheric perspective and spatial layout throughout the clip. @image2 is the character reference; preserve her identity, face, long pale-blonde hair, black outfit, body proportions and visible butterfly hair ornament consistently. + +Create a premium photorealistic fantasy-film shot in which the adult woman from @image2 stands in the field from @image1 while an enormous ancient dragon occupies the middle distance. Keep the tone majestic and cinematic rather than comedic or violent. The woman remains calm, confident and physically grounded; the dragon is powerful and irritated but believable, with detailed scales, leathery wings, articulated joints, natural breathing and convincing weight. Use warm late-afternoon sunlight, volumetric atmosphere, realistic shadows on the wheat and a restrained cinematic colour grade. No modern objects, logos, subtitles, watermarks or unexplained changes of wardrobe. + +[Shot 1] At 00:00.000, establish a wide view of the adult woman standing in the wheat field with the dragon behind and to one side. The dragon watches her with tense posture and slowly flexes its wings. Wheat moves in the wind and both subjects remain anatomically stable. + +[Shot 2] At 00:02.500, move to a medium over-the-shoulder composition from behind the woman toward the dragon. The dragon lowers its head, opens its mouth and gives a powerful roar without breathing fire. Dust and wheat respond subtly to the force of the breath; keep the woman composed. + +[Shot 3] At 00:05.000, cut to a closer side angle as the woman turns toward the dragon and raises one open palm in a clear stop gesture. The dragon leans closer, agitated but cautious. Maintain the same faces, hair, clothing, scale relationship and sunlight direction. + +[Shot 4] At 00:07.500, the woman makes one decisive, physically readable open-handed slap across the dragon's snout. Keep it stylised and non-graphic: no injury, blood or broken anatomy. The dragon recoils with a startled whimper and folds its head slightly. + +[Shot 5] At 00:09.000, return to a cinematic medium-wide view. The dragon is now meek and subdued while the woman turns back toward camera, spreads her arms slightly and gives a silent, incredulous expression. End with a natural breeze, settling wings and stable continuity through the final frame. + +overall_soundscape: Warm open-field wind, wheat brushing together, distant birds, heavy dragon breathing, wing fabric and joints moving, one powerful roar, the soft physical sound of the slap and a subdued dragon whimper. No dialogue, no screaming and no modern sound effects. + +non_diegetic_music: N/A. Keep the test free of music so image and motion quality can be compared directly. diff --git a/optimized/R2V/minimax/sparse-latent-upscale/references/Char034_TurnaroundFull_001.png b/optimized/R2V/minimax/sparse-latent-upscale/references/Char034_TurnaroundFull_001.png new file mode 100644 index 0000000..8c2c0fc Binary files /dev/null and b/optimized/R2V/minimax/sparse-latent-upscale/references/Char034_TurnaroundFull_001.png differ diff --git a/optimized/R2V/minimax/sparse-latent-upscale/references/Scene022.png b/optimized/R2V/minimax/sparse-latent-upscale/references/Scene022.png new file mode 100644 index 0000000..f5a86f5 Binary files /dev/null and b/optimized/R2V/minimax/sparse-latent-upscale/references/Scene022.png differ diff --git a/optimized/R2V/minimax/sparse-latent-upscale/variables.json b/optimized/R2V/minimax/sparse-latent-upscale/variables.json new file mode 100644 index 0000000..06e497f --- /dev/null +++ b/optimized/R2V/minimax/sparse-latent-upscale/variables.json @@ -0,0 +1,27 @@ +{ + "workflow": "sparse-latent-upscale", + "source_workflow": "workflow-source.json", + "fixed": { + "seed": 317182692766673, + "frames": 243, + "fps": 24, + "seconds": 10.125, + "prompt_file": "prompt.txt", + "initial_sampler": "euler", + "initial_scheduler": "beta", + "initial_steps": 8, + "initial_denoise": 1.0, + "refinement_sampler": "sa_solver", + "refinement_scheduler": "simple", + "refinement_steps": 3, + "refinement_denoise": 0.2, + "refinement_cfg": 1, + "reference_images": { + "@image1": "references/Scene022.png", + "@image2": "references/Char034_TurnaroundFull_001.png" + } + }, + "test_ids": [ + "SLU-01", "SLU-02", "SLU-03", "SLU-04", "SLU-05", "SLU-06", "SLU-07", "SLU-08", "SLU-09", "SLU-10", "SLU-11", "SLU-12", "SLU-13", "SLU-14", "SLU-15", "SLU-16", "SLU-17" + ] +} diff --git a/optimized/R2V/minimax/sparse-latent-upscale/workflow-source.json b/optimized/R2V/minimax/sparse-latent-upscale/workflow-source.json new file mode 100644 index 0000000..573e3b8 --- /dev/null +++ b/optimized/R2V/minimax/sparse-latent-upscale/workflow-source.json @@ -0,0 +1,573 @@ +{ + "119": { + "inputs": { + "vae_name": "minimax_h3_video_vae_int8_convrot.safetensors" + }, + "class_type": "VAELoader", + "_meta": { + "title": "Load VAE" + } + }, + "120": { + "inputs": { + "vae_name": "minimax_h3_audio_vae_fp32.safetensors" + }, + "class_type": "VAELoader", + "_meta": { + "title": "Load VAE" + } + }, + "123": { + "inputs": { + "sampler_name": "euler" + }, + "class_type": "KSamplerSelect", + "_meta": { + "title": "KSamplerSelect" + } + }, + "124": { + "inputs": { + "scheduler": "beta", + "steps": 8, + "denoise": 1, + "model": [ + "212", + 0 + ] + }, + "class_type": "BasicScheduler", + "_meta": { + "title": "BasicScheduler" + } + }, + "125": { + "inputs": { + "noise": [ + "129", + 0 + ], + "guider": [ + "126", + 0 + ], + "sampler": [ + "123", + 0 + ], + "sigmas": [ + "124", + 0 + ], + "latent_image": [ + "153", + 1 + ] + }, + "class_type": "SamplerCustomAdvanced", + "_meta": { + "title": "SamplerCustomAdvanced" + } + }, + "126": { + "inputs": { + "model": [ + "212", + 0 + ], + "conditioning": [ + "153", + 0 + ] + }, + "class_type": "BasicGuider", + "_meta": { + "title": "Basic Guider" + } + }, + "128": { + "inputs": { + "clip_name": "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors", + "type": "minimax", + "device": "default" + }, + "class_type": "CLIPLoader", + "_meta": { + "title": "Load CLIP" + } + }, + "129": { + "inputs": { + "noise_seed": 317182692766673 + }, + "class_type": "RandomNoise", + "_meta": { + "title": "RandomNoise" + } + }, + "149": { + "inputs": { + "fps": 24, + "preset": "H.264 (compatible)", + "filename_prefix": [ + "153", + 10 + ], + "save_output": true, + "save_metadata": true, + "fallback_on_failure": true, + "container": "mp4", + "images": [ + "186", + 0 + ], + "audio": [ + "186", + 1 + ] + }, + "class_type": "CSGlideVideoCS", + "_meta": { + "title": "Glide Video" + } + }, + "153": { + "inputs": { + "h3_data": "{\"mode\":\"ref2va\",\"width\":672,\"height\":384,\"length\":243,\"ref_image_size\":\"match\",\"ref_refine_scale\":1,\"pace\":2.5,\"prompt\":\"detailed_description:\\n@image1 @image2 are the available visual references.\\n\\n[Shot 1] At 00:00.000, @image2 is standing in the field from @image1. She is smiling and the Dragon seems angry, roaring and breathing fire.\\n[Shot 2] At 00:02.500, The dragon lowers its head towards @image2 and roars again.\\n[Shot 3] At 00:05.000, @image2 turns towards the dragon and with an open palm, raises her hand and with a large powerful swing, slaps the dragon hard across the face.\\n[Shot 4] At 00:07.500, The dragon whimpers and turns meek and timid.\\n[Shot 5] At 00:09.000, @image2 turns back to the camera, spreads his arms out wide and aggressively says \\\"What?!\\\".\\n\\n\\n\\noverall_soundscape: Desert wind, no voices required unless manually added.\\n\\nnon_diegetic_music: N/A.\",\"slots\":{\"first\":{},\"last\":{},\"images\":[{\"file\":\"cglide/Scene022.png\"},{\"file\":\"cglide/Gemma_TurnaroundFull_001.png\"},{},{},{},{},{},{},{}],\"videos\":[{},{},{}],\"audios\":[{},{},{}]},\"cont\":{},\"clip\":null}", + "clip": [ + "128", + 0 + ], + "vae": [ + "119", + 0 + ], + "audio_vae": [ + "120", + 0 + ] + }, + "class_type": "CSGlideCastCS", + "_meta": { + "title": "H3 Studio" + } + }, + "174": { + "inputs": { + "samples": [ + "125", + 0 + ], + "vae": [ + "120", + 0 + ] + }, + "class_type": "VAEDecodeAudio", + "_meta": { + "title": "VAE Decode Audio" + } + }, + "176": { + "inputs": { + "samples": [ + "236", + 0 + ], + "vae": [ + "119", + 0 + ] + }, + "class_type": "VAEDecode", + "_meta": { + "title": "VAE Decode" + } + }, + "186": { + "inputs": { + "source_video": [ + "153", + 7 + ], + "overlap_frames": [ + "153", + 6 + ], + "seam_mode": "early_cut", + "seam_blend_frames": 6, + "match_levels": true, + "fps": 24, + "match_tail": true, + "images": [ + "176", + 0 + ], + "audio": [ + "174", + 0 + ] + }, + "class_type": "CSGlideJoinCS", + "_meta": { + "title": "Glide Join" + } + }, + "204": { + "inputs": { + "attention": "comfy kitchen attention", + "model": [ + "220", + 0 + ] + }, + "class_type": "ModelAttentionBackend", + "_meta": { + "title": "Model Attention Backend" + } + }, + "205": { + "inputs": { + "head_chunks": 4, + "model": [ + "204", + 0 + ] + }, + "class_type": "MiniMaxLowVRAMAttention", + "_meta": { + "title": "MiniMax H3 Low VRAM Attention" + } + }, + "206": { + "inputs": { + "chunks": 2, + "seq_threshold": 4096, + "model": [ + "205", + 0 + ] + }, + "class_type": "MiniMaxChunkFeedForward", + "_meta": { + "title": "MiniMax H3 Chunk FeedForward" + } + }, + "212": { + "inputs": { + "max_resolution": 1024, + "jpeg_quality": 80, + "suppress_default_preview": true, + "preview_frames": 241, + "preview_fps": 7, + "tiny_vae": "taeh3.safetensors", + "model": [ + "216", + 0 + ] + }, + "class_type": "ModelPreviewOverrideKJ", + "_meta": { + "title": "Model Preview Override" + } + }, + "216": { + "inputs": { + "PowerLoraLoaderHeaderWidget": { + "type": "PowerLoraLoaderHeaderWidget" + }, + "lora_1": { + "on": true, + "lora": "minimax\\VBVR_H3_attn_only.safetensors", + "strength": 0.8 + }, + "lora_2": { + "on": true, + "lora": "minimax\\MysticXXX_MMH3-V4-ref2va.safetensors", + "strength": 0.45 + }, + "➕ Add Lora": "", + "model": [ + "206", + 0 + ] + }, + "class_type": "Power Lora Loader (rgthree)", + "_meta": { + "title": "Power Lora Loader (rgthree)" + } + }, + "217": { + "inputs": { + "text": "C:\\AI\\ComfyUI_windows_portable\\ComfyUI\\output\\glide_00007.mp4", + "anything": [ + "149", + 0 + ] + }, + "class_type": "easy showAnything", + "_meta": { + "title": "Show Any" + } + }, + "220": { + "inputs": { + "reuse_threshold": 0.3, + "start_percent": 0.2, + "end_percent": 0.9, + "verbose": false, + "model": [ + "250", + 0 + ] + }, + "class_type": "EasyCache", + "_meta": { + "title": "EasyCache" + } + }, + "236": { + "inputs": { + "cfg": 1, + "model": [ + "212", + 0 + ], + "conditioning": [ + "153", + 0 + ], + "latent": [ + "125", + 0 + ], + "noise": [ + "243", + 0 + ], + "sampler": [ + "242", + 0 + ], + "sigmas": [ + "241", + 0 + ], + "latent_upscale_param": [ + "238", + 0 + ], + "spatial_split_param": [ + "249", + 0 + ] + }, + "class_type": "MMH3UltimateUpscale", + "_meta": { + "title": "MMH3 Ultimate Upscale" + } + }, + "238": { + "inputs": { + "model_name": "minimax_h3_latent_upscaler_3d_conv_v1_fp16.safetensors", + "width": [ + "245", + 0 + ], + "height": [ + "245", + 1 + ], + "device": "cuda", + "precision": "fp16" + }, + "class_type": "MMH3LatentUpscaleWithModelParams", + "_meta": { + "title": "MMH3 Latent Upscale with Model Params" + } + }, + "241": { + "inputs": { + "scheduler": "simple", + "steps": 3, + "denoise": 0.2, + "model": [ + "212", + 0 + ] + }, + "class_type": "BasicScheduler", + "_meta": { + "title": "BasicScheduler" + } + }, + "242": { + "inputs": { + "sampler_name": "sa_solver" + }, + "class_type": "KSamplerSelect", + "_meta": { + "title": "KSamplerSelect" + } + }, + "243": { + "inputs": { + "noise_seed": 363742558835919 + }, + "class_type": "RandomNoise", + "_meta": { + "title": "RandomNoise" + } + }, + "245": { + "inputs": { + "ResolutionState": "{\"mode\":\"preset\",\"ratio\":\"16:9\",\"w\":1344,\"h\":768,\"custom_w\":1024,\"custom_h\":1024,\"custom_ratio_w\":4,\"custom_ratio_h\":3,\"snap\":16}" + }, + "class_type": "PixaromaResolution", + "_meta": { + "title": "Resolution Pixaroma" + } + }, + "247": { + "inputs": { + "text": { + "rows": [ + 0, + 22 + ], + "cols": [ + 0, + 46 + ], + "tile_rows": [ + 30, + 26 + ], + "tile_cols": [ + 54, + 38 + ], + "row_overlaps": [ + 0, + 8 + ], + "col_overlaps": [ + 0, + 8 + ], + "orig_h": 48, + "orig_w": 84, + "joint": true, + "steps": 3, + "tiles": 4, + "chunk": 0 + }, + "anything": [ + "236", + 2 + ] + }, + "class_type": "easy showAnything", + "_meta": { + "title": "Show Any" + } + }, + "248": { + "inputs": { + "text": { + "chunk": 0, + "frame_start": 0, + "frame_count": 243, + "video_tokens": [ + 0, + 72 + ], + "audio_tokens": [ + 0, + 405 + ], + "upscaled": true, + "spatial_h": 48, + "spatial_w": 84 + }, + "anything": [ + "236", + 1 + ] + }, + "class_type": "easy showAnything", + "_meta": { + "title": "Show Any" + } + }, + "249": { + "inputs": { + "upscale_width": 1024, + "upscale_height": [ + "260", + 0 + ], + "tile_size_mode": "specific_size", + "tile_width": 864, + "tile_height": [ + "260", + 1 + ], + "grid_rows": 2, + "grid_cols": 2, + "spatial_w_overlap": 128, + "spatial_h_overlap": 128, + "fade_width": 64, + "fade_height": 64, + "min_tile_size": 256, + "overlap_mode": "later", + "overlap_blend": "linear", + "joint_steps": 0.05, + "token_budget": 4096 + }, + "class_type": "MMH3SpatialSplitParams", + "_meta": { + "title": "MMH3 Spatial Split Params" + } + }, + "250": { + "inputs": { + "selection": "vsa", + "selection.keep_percent": 10, + "start_percent": 0.2, + "end_percent": 1, + "dense_blocks": "", + "min_tokens": 12288, + "extra_tokens": 256, + "sink_conditioning": "exact_kv_and_rows", + "verbose": false, + "model": [ + "251", + 0 + ] + }, + "class_type": "BlockSparseAttention", + "_meta": { + "title": "Model Sparse Attention" + } + }, + "251": { + "inputs": { + "unet_name": "fastvideo_fasth3_8step_v2_pruned_int8_convrot.safetensors", + "weight_dtype": "default" + }, + "class_type": "UNETLoader", + "_meta": { + "title": "Load Diffusion Model" + } + }, + "260": { + "inputs": { + "ResolutionState": "{\"mode\":\"preset\",\"ratio\":\"16:9\",\"w\":832,\"h\":480,\"custom_w\":1024,\"custom_h\":1024,\"custom_ratio_w\":4,\"custom_ratio_h\":3,\"snap\":16}" + }, + "class_type": "PixaromaResolution", + "_meta": { + "title": "Resolution Pixaroma" + } + } +} \ No newline at end of file diff --git a/optimized/R2V/minimax/sparse-latent-upscale/workflow.json b/optimized/R2V/minimax/sparse-latent-upscale/workflow.json new file mode 100644 index 0000000..b5c1837 --- /dev/null +++ b/optimized/R2V/minimax/sparse-latent-upscale/workflow.json @@ -0,0 +1,573 @@ +{ + "119": { + "inputs": { + "vae_name": "minimax_h3_video_vae_int8_convrot.safetensors" + }, + "class_type": "VAELoader", + "_meta": { + "title": "Load VAE" + } + }, + "120": { + "inputs": { + "vae_name": "minimax_h3_audio_vae_fp32.safetensors" + }, + "class_type": "VAELoader", + "_meta": { + "title": "Load VAE" + } + }, + "123": { + "inputs": { + "sampler_name": "euler" + }, + "class_type": "KSamplerSelect", + "_meta": { + "title": "KSamplerSelect" + } + }, + "124": { + "inputs": { + "scheduler": "beta", + "steps": 8, + "denoise": 1, + "model": [ + "212", + 0 + ] + }, + "class_type": "BasicScheduler", + "_meta": { + "title": "BasicScheduler" + } + }, + "125": { + "inputs": { + "noise": [ + "129", + 0 + ], + "guider": [ + "126", + 0 + ], + "sampler": [ + "123", + 0 + ], + "sigmas": [ + "124", + 0 + ], + "latent_image": [ + "153", + 1 + ] + }, + "class_type": "SamplerCustomAdvanced", + "_meta": { + "title": "SamplerCustomAdvanced" + } + }, + "126": { + "inputs": { + "model": [ + "212", + 0 + ], + "conditioning": [ + "153", + 0 + ] + }, + "class_type": "BasicGuider", + "_meta": { + "title": "Basic Guider" + } + }, + "128": { + "inputs": { + "clip_name": "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors", + "type": "minimax", + "device": "default" + }, + "class_type": "CLIPLoader", + "_meta": { + "title": "Load CLIP" + } + }, + "129": { + "inputs": { + "noise_seed": 317182692766673 + }, + "class_type": "RandomNoise", + "_meta": { + "title": "RandomNoise" + } + }, + "149": { + "inputs": { + "fps": 24, + "preset": "H.264 (compatible)", + "filename_prefix": [ + "153", + 10 + ], + "save_output": true, + "save_metadata": true, + "fallback_on_failure": true, + "container": "mp4", + "images": [ + "186", + 0 + ], + "audio": [ + "186", + 1 + ] + }, + "class_type": "CSGlideVideoCS", + "_meta": { + "title": "Glide Video" + } + }, + "153": { + "inputs": { + "h3_data": "{\"mode\":\"ref2va\",\"width\":672,\"height\":384,\"length\":243,\"ref_image_size\":\"match\",\"ref_refine_scale\":1,\"pace\":2.5,\"prompt\":\"detailed_description:\\n@image1 is the environment and master scene reference. Preserve the wheat field, distant tree line, daylight, atmospheric perspective and spatial layout throughout the clip. @image2 is the character reference; preserve her identity, face, long pale-blonde hair, black outfit, body proportions and visible butterfly hair ornament consistently.\\n\\nCreate a premium photorealistic fantasy-film shot in which the adult woman from @image2 stands in the field from @image1 while an enormous ancient dragon occupies the middle distance. Keep the tone majestic and cinematic rather than comedic or violent. The woman remains calm, confident and physically grounded; the dragon is powerful and irritated but believable, with detailed scales, leathery wings, articulated joints, natural breathing and convincing weight. Use warm late-afternoon sunlight, volumetric atmosphere, realistic shadows on the wheat and a restrained cinematic colour grade. No modern objects, logos, subtitles, watermarks or unexplained changes of wardrobe.\\n\\n[Shot 1] At 00:00.000, establish a wide view of the adult woman standing in the wheat field with the dragon behind and to one side. The dragon watches her with tense posture and slowly flexes its wings. Wheat moves in the wind and both subjects remain anatomically stable.\\n\\n[Shot 2] At 00:02.500, move to a medium over-the-shoulder composition from behind the woman toward the dragon. The dragon lowers its head, opens its mouth and gives a powerful roar without breathing fire. Dust and wheat respond subtly to the force of the breath; keep the woman composed.\\n\\n[Shot 3] At 00:05.000, cut to a closer side angle as the woman turns toward the dragon and raises one open palm in a clear stop gesture. The dragon leans closer, agitated but cautious. Maintain the same faces, hair, clothing, scale relationship and sunlight direction.\\n\\n[Shot 4] At 00:07.500, the woman makes one decisive, physically readable open-handed slap across the dragon's snout. Keep it stylised and non-graphic: no injury, blood or broken anatomy. The dragon recoils with a startled whimper and folds its head slightly.\\n\\n[Shot 5] At 00:09.000, return to a cinematic medium-wide view. The dragon is now meek and subdued while the woman turns back toward camera, spreads her arms slightly and gives a silent, incredulous expression. End with a natural breeze, settling wings and stable continuity through the final frame.\\n\\noverall_soundscape: Warm open-field wind, wheat brushing together, distant birds, heavy dragon breathing, wing fabric and joints moving, one powerful roar, the soft physical sound of the slap and a subdued dragon whimper. No dialogue, no screaming and no modern sound effects.\\n\\nnon_diegetic_music: N/A. Keep the test free of music so image and motion quality can be compared directly.\\n\",\"slots\":{\"first\":{},\"last\":{},\"images\":[{\"file\":\"cglide/Scene022.png\"},{\"file\":\"cglide/Char034_TurnaroundFull_001.png\"},{},{},{},{},{},{},{}],\"videos\":[{},{},{}],\"audios\":[{},{},{}]},\"cont\":{},\"clip\":null}", + "clip": [ + "128", + 0 + ], + "vae": [ + "119", + 0 + ], + "audio_vae": [ + "120", + 0 + ] + }, + "class_type": "CSGlideCastCS", + "_meta": { + "title": "H3 Studio" + } + }, + "174": { + "inputs": { + "samples": [ + "125", + 0 + ], + "vae": [ + "120", + 0 + ] + }, + "class_type": "VAEDecodeAudio", + "_meta": { + "title": "VAE Decode Audio" + } + }, + "176": { + "inputs": { + "samples": [ + "236", + 0 + ], + "vae": [ + "119", + 0 + ] + }, + "class_type": "VAEDecode", + "_meta": { + "title": "VAE Decode" + } + }, + "186": { + "inputs": { + "source_video": [ + "153", + 7 + ], + "overlap_frames": [ + "153", + 6 + ], + "seam_mode": "early_cut", + "seam_blend_frames": 6, + "match_levels": true, + "fps": 24, + "match_tail": true, + "images": [ + "176", + 0 + ], + "audio": [ + "174", + 0 + ] + }, + "class_type": "CSGlideJoinCS", + "_meta": { + "title": "Glide Join" + } + }, + "204": { + "inputs": { + "attention": "comfy kitchen attention", + "model": [ + "220", + 0 + ] + }, + "class_type": "ModelAttentionBackend", + "_meta": { + "title": "Model Attention Backend" + } + }, + "205": { + "inputs": { + "head_chunks": 4, + "model": [ + "204", + 0 + ] + }, + "class_type": "MiniMaxLowVRAMAttention", + "_meta": { + "title": "MiniMax H3 Low VRAM Attention" + } + }, + "206": { + "inputs": { + "chunks": 2, + "seq_threshold": 4096, + "model": [ + "205", + 0 + ] + }, + "class_type": "MiniMaxChunkFeedForward", + "_meta": { + "title": "MiniMax H3 Chunk FeedForward" + } + }, + "212": { + "inputs": { + "max_resolution": 1024, + "jpeg_quality": 80, + "suppress_default_preview": true, + "preview_frames": 241, + "preview_fps": 7, + "tiny_vae": "taeh3.safetensors", + "model": [ + "216", + 0 + ] + }, + "class_type": "ModelPreviewOverrideKJ", + "_meta": { + "title": "Model Preview Override" + } + }, + "216": { + "inputs": { + "PowerLoraLoaderHeaderWidget": { + "type": "PowerLoraLoaderHeaderWidget" + }, + "lora_1": { + "on": true, + "lora": "minimax\\VBVR_H3_attn_only.safetensors", + "strength": 0.8 + }, + "lora_2": { + "on": true, + "lora": "minimax\\MysticXXX_MMH3-V4-ref2va.safetensors", + "strength": 0.45 + }, + "➕ Add Lora": "", + "model": [ + "206", + 0 + ] + }, + "class_type": "Power Lora Loader (rgthree)", + "_meta": { + "title": "Power Lora Loader (rgthree)" + } + }, + "217": { + "inputs": { + "text": "C:\\AI\\ComfyUI_windows_portable\\ComfyUI\\output\\glide_00007.mp4", + "anything": [ + "149", + 0 + ] + }, + "class_type": "easy showAnything", + "_meta": { + "title": "Show Any" + } + }, + "220": { + "inputs": { + "reuse_threshold": 0.3, + "start_percent": 0.2, + "end_percent": 0.9, + "verbose": false, + "model": [ + "250", + 0 + ] + }, + "class_type": "EasyCache", + "_meta": { + "title": "EasyCache" + } + }, + "236": { + "inputs": { + "cfg": 1, + "model": [ + "212", + 0 + ], + "conditioning": [ + "153", + 0 + ], + "latent": [ + "125", + 0 + ], + "noise": [ + "243", + 0 + ], + "sampler": [ + "242", + 0 + ], + "sigmas": [ + "241", + 0 + ], + "latent_upscale_param": [ + "238", + 0 + ], + "spatial_split_param": [ + "249", + 0 + ] + }, + "class_type": "MMH3UltimateUpscale", + "_meta": { + "title": "MMH3 Ultimate Upscale" + } + }, + "238": { + "inputs": { + "model_name": "minimax_h3_latent_upscaler_3d_conv_v1_fp16.safetensors", + "width": [ + "245", + 0 + ], + "height": [ + "245", + 1 + ], + "device": "cuda", + "precision": "fp16" + }, + "class_type": "MMH3LatentUpscaleWithModelParams", + "_meta": { + "title": "MMH3 Latent Upscale with Model Params" + } + }, + "241": { + "inputs": { + "scheduler": "simple", + "steps": 3, + "denoise": 0.2, + "model": [ + "212", + 0 + ] + }, + "class_type": "BasicScheduler", + "_meta": { + "title": "BasicScheduler" + } + }, + "242": { + "inputs": { + "sampler_name": "sa_solver" + }, + "class_type": "KSamplerSelect", + "_meta": { + "title": "KSamplerSelect" + } + }, + "243": { + "inputs": { + "noise_seed": 363742558835919 + }, + "class_type": "RandomNoise", + "_meta": { + "title": "RandomNoise" + } + }, + "245": { + "inputs": { + "ResolutionState": "{\"mode\":\"preset\",\"ratio\":\"16:9\",\"w\":1344,\"h\":768,\"custom_w\":1024,\"custom_h\":1024,\"custom_ratio_w\":4,\"custom_ratio_h\":3,\"snap\":16}" + }, + "class_type": "PixaromaResolution", + "_meta": { + "title": "Resolution Pixaroma" + } + }, + "247": { + "inputs": { + "text": { + "rows": [ + 0, + 22 + ], + "cols": [ + 0, + 46 + ], + "tile_rows": [ + 30, + 26 + ], + "tile_cols": [ + 54, + 38 + ], + "row_overlaps": [ + 0, + 8 + ], + "col_overlaps": [ + 0, + 8 + ], + "orig_h": 48, + "orig_w": 84, + "joint": true, + "steps": 3, + "tiles": 4, + "chunk": 0 + }, + "anything": [ + "236", + 2 + ] + }, + "class_type": "easy showAnything", + "_meta": { + "title": "Show Any" + } + }, + "248": { + "inputs": { + "text": { + "chunk": 0, + "frame_start": 0, + "frame_count": 243, + "video_tokens": [ + 0, + 72 + ], + "audio_tokens": [ + 0, + 405 + ], + "upscaled": true, + "spatial_h": 48, + "spatial_w": 84 + }, + "anything": [ + "236", + 1 + ] + }, + "class_type": "easy showAnything", + "_meta": { + "title": "Show Any" + } + }, + "249": { + "inputs": { + "upscale_width": 1024, + "upscale_height": [ + "260", + 0 + ], + "tile_size_mode": "specific_size", + "tile_width": 864, + "tile_height": [ + "260", + 1 + ], + "grid_rows": 2, + "grid_cols": 2, + "spatial_w_overlap": 128, + "spatial_h_overlap": 128, + "fade_width": 64, + "fade_height": 64, + "min_tile_size": 256, + "overlap_mode": "later", + "overlap_blend": "linear", + "joint_steps": 0.05, + "token_budget": 4096 + }, + "class_type": "MMH3SpatialSplitParams", + "_meta": { + "title": "MMH3 Spatial Split Params" + } + }, + "250": { + "inputs": { + "selection": "vsa", + "selection.keep_percent": 10, + "start_percent": 0.2, + "end_percent": 1, + "dense_blocks": "", + "min_tokens": 12288, + "extra_tokens": 256, + "sink_conditioning": "exact_kv_and_rows", + "verbose": false, + "model": [ + "251", + 0 + ] + }, + "class_type": "BlockSparseAttention", + "_meta": { + "title": "Model Sparse Attention" + } + }, + "251": { + "inputs": { + "unet_name": "fastvideo_fasth3_8step_v2_pruned_int8_convrot.safetensors", + "weight_dtype": "default" + }, + "class_type": "UNETLoader", + "_meta": { + "title": "Load Diffusion Model" + } + }, + "260": { + "inputs": { + "ResolutionState": "{\"mode\":\"preset\",\"ratio\":\"16:9\",\"w\":832,\"h\":480,\"custom_w\":1024,\"custom_h\":1024,\"custom_ratio_w\":4,\"custom_ratio_h\":3,\"snap\":16}" + }, + "class_type": "PixaromaResolution", + "_meta": { + "title": "Resolution Pixaroma" + } + } +}