diff --git a/optimized/R2V/minimax/continuum-tiled/README.md b/optimized/R2V/minimax/continuum-tiled/README.md new file mode 100644 index 0000000..804988c --- /dev/null +++ b/optimized/R2V/minimax/continuum-tiled/README.md @@ -0,0 +1,75 @@ +# MiniMax H3 Continuum — tiled upscale / multi-clip prompt study + +This folder contains the supplied MiniMax H3 Continuum workflow with the +DUMAS tiled latent upscale second pass. It is the working candidate for +testing prompt direction and multi-clip construction toward a **20-second +final video**. + +## Source and status + +- Source: `Unsaved_Workflow_5_1.json`, supplied 2026-10-14. +- Status: imported baseline; smoke validation pending on the live server. +- Workflow: [`workflow.json`](workflow.json) +- References: [`character-reference.png`](character-reference.png) and + [`scene-reference.png`](scene-reference.png) +- Custom nodes: DUMAS H3 Continuum, DUMAS H3 Continuum Tiled, and the native + MiniMax H3/Fearnworks/PlagueKind dependencies used by the graph. + +## What is fixed + +These values are held constant while testing prompts and clip scaling: + +- Model: `fastvideo_fasth3_8step_v2_pruned_int8_convrot.safetensors` +- Text encoder: `qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors` +- Video VAE: `minimax_h3_video_vae_int8_convrot.safetensors` +- Audio VAE: `minimax_h3_audio_vae_fp32.safetensors` +- Main sampler: `euler_ancestral` +- Main scheduler: `simple` +- Main steps: `8` +- Main denoise: `1.0` +- Main seed: `12` +- Tiled second-pass sampler: `sa_solver` +- Tiled second-pass scheduler: `simple` +- Tiled second-pass steps: `2` +- Tiled second-pass denoise: `0.2` +- Tiled second-pass CFG: `1.0` +- Final output: `1920×1088` +- Latent upscale model: `minimax_h3_latent_upscaler_3d_conv_v1_fp16.safetensors` +- Tile geometry: `1056×640`, 96-pixel overlap, 32-pixel fade +- Tiled mode: enabled through `DUMAS H3 Continuum Second Pass` + +The imported graph starts at two 4-second Continuum chunks. The test matrix +in [`tests.json`](tests.json) increases the number of chunks until the +20-second target is reached. No sampler, scheduler, CFG, upscale or tile +changes are included in this study. + +## Prompt strategy + +The same character and scene references, seed, camera language, and visual +direction are retained across the matrix. Prompt tests change only the +structure and specificity of the action plan: + +1. Compact chronological action beats. +2. Detailed chronological beats with explicit continuity. +3. Detailed beats with camera direction and physical state tracking. +4. Full 20-second plan with five linked 4-second beats. + +## Test procedure + +1. Run the 2-chunk smoke row first. +2. Confirm the final output is routed through the DUMAS Second Pass with + `tiled_upscale=true` and reaches 1920×1088. +3. Repeat with the next prompt variant using the same 2 chunks. +4. Keep the best prompt fixed, then test 3, 4, and 5 chunks. +5. Stop if the server errors, OOMs, or runtime growth becomes disproportionate. +6. Record each result beside the matrix with prompt hash, chunk count, + runtime, output dimensions, and a motion/continuity verdict. + +## Files + +- `workflow.json` — supplied ComfyUI API graph. +- `prompt.txt` — controlled base prompt for the study. +- `variables.json` — immutable constants and allowed test variables. +- `tests.json` — ordered test matrix. +- `character-reference.png`, `scene-reference.png` — fixed references. +- `results/` — reserved for labelled outputs and metadata. diff --git a/optimized/R2V/minimax/continuum-tiled/character-reference.png b/optimized/R2V/minimax/continuum-tiled/character-reference.png new file mode 100644 index 0000000..c5eca09 Binary files /dev/null and b/optimized/R2V/minimax/continuum-tiled/character-reference.png differ diff --git a/optimized/R2V/minimax/continuum-tiled/prompt.txt b/optimized/R2V/minimax/continuum-tiled/prompt.txt new file mode 100644 index 0000000..7d39222 --- /dev/null +++ b/optimized/R2V/minimax/continuum-tiled/prompt.txt @@ -0,0 +1,36 @@ +@R1 is the environment reference and @R2 is the character reference. Preserve +the character's identity, clothing, proportions, and visible features. Preserve +the environment's layout, lighting, atmosphere, and geography. Use realistic +physical motion, stable anatomy, coherent screen direction, and continuous +state between every beat. No dialogue unless explicitly stated; use natural +ambient sound and restrained cinematic music. + +[0-4s] @R2 enters the environment from the left and walks forward with alert +curiosity, studying the surroundings. The camera tracks backwards in a smooth +medium shot. A sudden movement in the distance makes the character stop and +turn toward the right. + +[4-8s] Continue seamlessly. @R2 walks toward the source of the movement while +the camera moves into a side-tracking shot. The character reaches a narrow +opening, looks through it, and raises one hand to signal caution. + +[8-12s] Continue with the same body position and direction. A gust of wind +disturbs loose environmental details. @R2 shields their face, steps sideways, +then notices a small object on the ground and carefully crouches to inspect it. + +[12-16s] @R2 stands while holding the object, looks from it to the distance, +and begins moving more quickly. The camera arcs around from profile to a frontal +three-quarter view. Keep the object in the same hand and preserve all clothing +and environmental continuity. + +[16-20s] @R2 reaches a clear vantage point and turns toward the camera as the +environment reveals a new distant event. The character reacts with focused +surprise, takes one controlled step back, then holds a ready stance while the +camera makes a slow push-in. End with natural motion, not a frozen pose. + +overall_soundscape: continuous environmental ambience matching the reference +scene, footsteps, clothing movement, wind, the object being handled, and subtle +physical reactions. No speech. + +non_diegetic_music: restrained cinematic tension that develops gradually and +remains subordinate to the soundscape. diff --git a/optimized/R2V/minimax/continuum-tiled/scene-reference.png b/optimized/R2V/minimax/continuum-tiled/scene-reference.png new file mode 100644 index 0000000..01ba2e3 Binary files /dev/null and b/optimized/R2V/minimax/continuum-tiled/scene-reference.png differ diff --git a/optimized/R2V/minimax/continuum-tiled/tests.json b/optimized/R2V/minimax/continuum-tiled/tests.json new file mode 100644 index 0000000..b727e00 --- /dev/null +++ b/optimized/R2V/minimax/continuum-tiled/tests.json @@ -0,0 +1,26 @@ +{ + "constants": { + "seed": 12, + "initial_resolution": "0.30 MP, 16:9, 32-pixel multiple", + "final_resolution": [1920, 1088], + "main_sampler": "euler_ancestral", + "main_scheduler": "simple", + "main_steps": 8, + "main_denoise": 1.0, + "upscale_sampler": "sa_solver", + "upscale_scheduler": "simple", + "upscale_steps": 2, + "upscale_denoise": 0.2, + "upscale_cfg": 1.0, + "tile": "1056x640 with 96px overlap and 32px fade", + "references": ["character-reference.png", "scene-reference.png"] + }, + "tests": [ + {"id":"CT-01","purpose":"2-chunk compact prompt smoke test","prompt_variant":"compact","chunks":2,"chunk_seconds":4,"target_duration_seconds":8,"status":"planned"}, + {"id":"CT-02","purpose":"2-chunk detailed continuity prompt","prompt_variant":"detailed-continuity","chunks":2,"chunk_seconds":4,"target_duration_seconds":8,"status":"planned"}, + {"id":"CT-03","purpose":"2-chunk explicit camera/action prompt","prompt_variant":"cinematic-camera","chunks":2,"chunk_seconds":4,"target_duration_seconds":8,"status":"planned"}, + {"id":"CT-04","purpose":"3 chunks using the best smoke prompt","prompt_variant":"best-of-smoke","chunks":3,"chunk_seconds":4,"target_duration_seconds":12,"status":"planned"}, + {"id":"CT-05","purpose":"4 chunks using the best prompt","prompt_variant":"best-of-smoke","chunks":4,"chunk_seconds":4,"target_duration_seconds":16,"status":"planned"}, + {"id":"CT-06","purpose":"5 chunks, full 20-second target","prompt_variant":"best-of-smoke","chunks":5,"chunk_seconds":4,"target_duration_seconds":20,"status":"planned"} + ] +} diff --git a/optimized/R2V/minimax/continuum-tiled/variables.json b/optimized/R2V/minimax/continuum-tiled/variables.json new file mode 100644 index 0000000..8a75689 --- /dev/null +++ b/optimized/R2V/minimax/continuum-tiled/variables.json @@ -0,0 +1,29 @@ +{ + "workflow": "workflow.json", + "prompt": "prompt.txt", + "references": ["character-reference.png", "scene-reference.png"], + "fixed": { + "model": "fastvideo_fasth3_8step_v2_pruned_int8_convrot.safetensors", + "text_encoder": "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors", + "video_vae": "minimax_h3_video_vae_int8_convrot.safetensors", + "audio_vae": "minimax_h3_audio_vae_fp32.safetensors", + "main_sampler": "euler_ancestral", + "main_scheduler": "simple", + "main_steps": 8, + "main_denoise": 1.0, + "seed": 12, + "upscale_sampler": "sa_solver", + "upscale_scheduler": "simple", + "upscale_steps": 2, + "upscale_denoise": 0.2, + "upscale_cfg": 1.0, + "upscale_resolution": [1920, 1088], + "tile_width": 1056, + "tile_height": 640, + "tile_overlap": [96, 96], + "tile_fade": [32, 32], + "latent_upscaler": "minimax_h3_latent_upscaler_3d_conv_v1_fp16.safetensors" + }, + "allowed_test_variables": ["prompt_variant", "chunks", "chunk_seconds"], + "target_duration_seconds": 20 +} diff --git a/optimized/R2V/minimax/continuum-tiled/workflow.json b/optimized/R2V/minimax/continuum-tiled/workflow.json new file mode 100644 index 0000000..806fca0 --- /dev/null +++ b/optimized/R2V/minimax/continuum-tiled/workflow.json @@ -0,0 +1,551 @@ +{ + "135": { + "inputs": { + "vae_name": "minimax_h3_video_vae_int8_convrot.safetensors" + }, + "class_type": "VAELoader", + "_meta": { + "title": "Video VAE" + } + }, + "136": { + "inputs": { + "vae_name": "minimax_h3_audio_vae_fp32.safetensors" + }, + "class_type": "VAELoader", + "_meta": { + "title": "Audio VAE" + } + }, + "139": { + "inputs": { + "sampler_name": "euler_ancestral" + }, + "class_type": "KSamplerSelect", + "_meta": { + "title": "Sampler" + } + }, + "142": { + "inputs": { + "clip_name": "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors", + "type": "minimax", + "device": "default" + }, + "class_type": "CLIPLoader", + "_meta": { + "title": "MiniMax H3 Text Encoder" + } + }, + "147": { + "inputs": { + "unet_name": "fastvideo_fasth3_8step_v2_pruned_int8_convrot.safetensors", + "weight_dtype": "default" + }, + "class_type": "UNETLoader", + "_meta": { + "title": "H3 Model" + } + }, + "150": { + "inputs": { + "enabled": false, + "blend_weight": 0.51, + "degree": 1, + "ridge_lambda": 0.1, + "window_size": 2, + "flex_window": 0.75, + "warmup_steps": 1, + "tail_actual_steps": 1, + "max_history": 8, + "debug": false, + "history_storage": "vram", + "bootstrap_first_forecast": true, + "anchor_residual_feedback": false, + "selective_rollback_correction": false, + "offline_smoothing_replay": true, + "audio_blend_weight": 0, + "offline_archive_storage": "system_ram", + "model_aware_mode": "off", + "model_aware_risk_threshold": 0.65, + "model_aware_trust_shrinkage": false, + "model_aware_replay_generic_correction": false, + "generic_correction_mode": "coordinate_rls", + "generic_correction_limiter": "hard_clip", + "generic_correction_limit": 0.4, + "generic_correction_attenuation": "no_attenuation", + "sa_pece_forecast_policy": "balanced", + "model": [ + "286", + 0 + ] + }, + "class_type": "SpectrumApplyMiniMaxH3", + "_meta": { + "title": "Spectrum" + } + }, + "153": { + "inputs": { + "scheduler": "simple", + "steps": 8, + "denoise": 1, + "model": [ + "150", + 0 + ] + }, + "class_type": "BasicScheduler", + "_meta": { + "title": "Scheduler" + } + }, + "188": { + "inputs": { + "value": "@R1 is the environment reference and @R2 is the character reference. Preserve identity, clothing, proportions, lighting, geography, and physical continuity. Use realistic motion and continuous screen direction.\n\n[0-4s] @R2 enters from the left and walks forward with alert curiosity. The camera tracks backwards in a smooth medium shot. A sudden movement makes the character stop and turn right.\n\n[4-8s] Continue seamlessly. @R2 approaches the movement while the camera side-tracks. The character reaches a narrow opening, looks through it, and raises one hand to signal caution.\n\n[8-12s] Continue with the same body position and direction. Wind disturbs the environment. @R2 shields their face, steps sideways, then crouches to inspect a small object.\n\n[12-16s] @R2 stands holding the object and moves more quickly. The camera arcs from profile to frontal three-quarter view. Preserve the object, clothing, and environment.\n\n[16-20s] @R2 reaches a vantage point and reacts with focused surprise to a distant event. The camera slowly pushes in as the character holds a ready stance.\n\noverall_soundscape: continuous environmental ambience, footsteps, wind, clothing movement, object handling, and subtle physical reactions. No speech.\n\nnon_diegetic_music: restrained cinematic tension.\n" + }, + "class_type": "PrimitiveStringMultiline", + "_meta": { + "title": "Text (Multiline)" + } + }, + "249": { + "inputs": { + "source": [ + "312", + 3 + ] + }, + "class_type": "PreviewAny", + "_meta": { + "title": "Preview as Text" + } + }, + "286": { + "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": 1 + }, + "➕ Add Lora": "", + "model": [ + "358", + 0 + ] + }, + "class_type": "Power Lora Loader (rgthree)", + "_meta": { + "title": "Power Lora Loader (rgthree)" + } + }, + "306": { + "inputs": { + "exact_total_duration": true, + "audio_seam": "Auto", + "video_seam": "Auto", + "buffer_backend": "Auto", + "diagnostics": "Basic", + "images": [ + "344", + 0 + ], + "audio": [ + "344", + 1 + ], + "assembly_plan": [ + "379", + 2 + ], + "driving_audio": [ + "312", + 4 + ] + }, + "class_type": "H3ContinuumAssembleSeamV35", + "_meta": { + "title": "DUMAS H3 Continuum Finalize" + } + }, + "312": { + "inputs": { + "prompt_mode": "Auto", + "chunks": 2, + "chunk_seconds": 4, + "aspect": "Auto from First Image", + "preset": "Draft — 0.30 MP", + "custom_mp": 0.3, + "continuity": "Balanced — 22 frames", + "base_seed": 12, + "audio_continuity": true, + "diagnostics": "Basic", + "reroll_from_chunk": "Auto", + "reroll_nonce": 0, + "strict_compatibility": false, + "debug": false, + "show_preview": true, + "run_storage": "Save + Auto Resume", + "run_name": "Continuum", + "reference_size": "Match Output", + "project_id": "b875c81a-ffa0-447c-894d-ffbc684fd3ea", + "video_reference_size": "Efficient - 0.4 MP", + "continuation_backend": "Standard", + "generation_mode": "Review Each Chunk", + "review_action": "Continue / Next", + "take_group": 0, + "take_revision_id": "", + "take_action": "Automatic", + "size_source": "Manual", + "width": [ + "359", + 0 + ], + "height": [ + "359", + 1 + ], + "video_reference_mode": "Repeat Reference", + "model": [ + "150", + 0 + ], + "clip": [ + "142", + 0 + ], + "video_vae": [ + "135", + 0 + ], + "sampler": [ + "139", + 0 + ], + "sigmas": [ + "153", + 0 + ], + "sequence_prompt": [ + "188", + 0 + ], + "audio_vae": [ + "136", + 0 + ], + "audio_references": [ + "331", + 0 + ], + "reference_images": [ + "348", + 0 + ] + }, + "class_type": "H3ContinuumSamplerV39", + "_meta": { + "title": "DUMAS H3 Continuum Sampler V3.9" + } + }, + "314": { + "inputs": { + "image": "character-reference.png" + }, + "class_type": "H3EasyLoadImage", + "_meta": { + "title": "DUMAS H3 Continuum Load Image" + } + }, + "316": { + "inputs": { + "image": "scene-reference.png" + }, + "class_type": "H3EasyLoadImage", + "_meta": { + "title": "DUMAS H3 Continuum Load Image" + } + }, + "331": { + "inputs": { + "reference_audio_vae": [ + "136", + 0 + ] + }, + "class_type": "H3ContinuumReferenceAudios", + "_meta": { + "title": "DUMAS H3 Continuum Reference Audios" + } + }, + "344": { + "inputs": { + "cache_mode": "Auto", + "ram_budget_mb": 256, + "disk_budget_gb": 8, + "reset_token": 0, + "video_samples": [ + "379", + 0 + ], + "video_vae": [ + "135", + 0 + ], + "audio_samples": [ + "379", + 1 + ], + "audio_vae": [ + "136", + 0 + ] + }, + "class_type": "H3DecodeCacheHelper", + "_meta": { + "title": "DUMAS Decode Cache Helper" + } + }, + "348": { + "inputs": { + "reference_use": "All chunks", + "reference_r1_chunks": "off", + "reference_r2_chunks": "off", + "reference_r3_chunks": "off", + "reference_r4_chunks": "off", + "reference_r5_chunks": "off", + "reference_r6_chunks": "off", + "reference_r7_chunks": "off", + "reference_r8_chunks": "off", + "reference_r9_chunks": "off", + "reference_image_1": [ + "316", + 0 + ], + "reference_image_2": [ + "314", + 0 + ] + }, + "class_type": "H3ContinuumReferenceImagesV39", + "_meta": { + "title": "DUMAS H3 Continuum Reference Images V3.9" + } + }, + "358": { + "inputs": { + "max_resolution": 1024, + "jpeg_quality": 80, + "suppress_default_preview": true, + "preview_frames": 240, + "preview_fps": 12, + "tiny_vae": "taeh3.safetensors", + "model": [ + "360", + 0 + ] + }, + "class_type": "ModelPreviewOverrideKJ", + "_meta": { + "title": "Model Preview Override" + } + }, + "359": { + "inputs": { + "aspect_ratio": "16:9 (Widescreen)", + "megapixels": 0.3, + "multiple": 32 + }, + "class_type": "ResolutionSelector", + "_meta": { + "title": "Resolution Selector" + } + }, + "360": { + "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": [ + "147", + 0 + ] + }, + "class_type": "BlockSparseAttention", + "_meta": { + "title": "Model Sparse Attention" + } + }, + "365": { + "inputs": { + "model_name": "minimax_h3_latent_upscaler_3d_conv_v1_fp16.safetensors", + "width": [ + "368", + 0 + ], + "height": [ + "368", + 1 + ], + "device": "cuda", + "precision": "fp16" + }, + "class_type": "MMH3LatentUpscaleWithModelParams", + "_meta": { + "title": "MMH3 Latent Upscale with Model Params" + } + }, + "366": { + "inputs": { + "upscale_width": 1024, + "upscale_height": 1024, + "tile_size_mode": "specific_size", + "tile_width": 1056, + "tile_height": 640, + "grid_rows": 2, + "grid_cols": 2, + "spatial_w_overlap": 96, + "spatial_h_overlap": 96, + "fade_width": 32, + "fade_height": 32, + "min_tile_size": 256, + "overlap_mode": "earlier", + "overlap_blend": "linear", + "joint_steps": true, + "token_budget": 70000 + }, + "class_type": "MMH3SpatialSplitParams", + "_meta": { + "title": "MMH3 Spatial Split Params" + } + }, + "368": { + "inputs": { + "ResolutionState": "{\"mode\":\"preset\",\"ratio\":\"16:9\",\"w\":1920,\"h\":1088,\"custom_w\":1024,\"custom_h\":1024,\"custom_ratio_w\":4,\"custom_ratio_h\":3,\"snap\":16}" + }, + "class_type": "PixaromaResolution", + "_meta": { + "title": "Resolution Pixaroma" + } + }, + "379": { + "inputs": { + "refine_seed": 2828258776954, + "tiled_upscale": true, + "upscale_sampler": "sa_solver", + "upscale_scheduler": "simple", + "upscale_steps": 2, + "upscale_denoise": 0.2, + "upscale_cfg": 1, + "upscale_noise_seed": 12, + "model": [ + "150", + 0 + ], + "clip": [ + "142", + 0 + ], + "sampler": [ + "381", + 0 + ], + "sigmas": [ + "380", + 0 + ], + "video_latents": [ + "312", + 0 + ], + "audio_latents": [ + "312", + 1 + ], + "assembly_plan": [ + "312", + 2 + ], + "refine_context": [ + "312", + 5 + ], + "latent_upscale_param": [ + "365", + 0 + ], + "spatial_split_param": [ + "366", + 0 + ] + }, + "class_type": "H3ContinuumSecondPassV35", + "_meta": { + "title": "DUMAS H3 Continuum Second Pass" + } + }, + "380": { + "inputs": { + "scheduler": "simple", + "steps": 2, + "denoise": 0.2, + "model": [ + "150", + 0 + ] + }, + "class_type": "BasicScheduler", + "_meta": { + "title": "BasicScheduler" + } + }, + "381": { + "inputs": { + "sampler_name": "sa_solver" + }, + "class_type": "KSamplerSelect", + "_meta": { + "title": "KSamplerSelect" + } + }, + "382": { + "inputs": { + "frame_rate": 24, + "loop_count": 0, + "filename_prefix": "AnimateDiff", + "format": "video/h264-mp4", + "pix_fmt": "yuv420p", + "crf": 19, + "save_metadata": true, + "trim_to_audio": false, + "pingpong": false, + "save_output": true, + "images": [ + "306", + 0 + ], + "audio": [ + "306", + 1 + ] + }, + "class_type": "VHS_VideoCombine", + "_meta": { + "title": "Video Combine 🎥🅥🅗🅢" + } + } +}