Queue connected squirrel continuation test

This commit is contained in:
Morpheus
2026-09-28 13:56:08 +00:00
parent 94c52900b9
commit ab17ec6f9b
3 changed files with 9 additions and 7 deletions
+1
View File
@@ -120,6 +120,7 @@ their row explicitly overrides resolution or duration.
| MM-R2V-66 | N cache | SLA 0.85 / block 32 + EasyCache defaults (reuse 0.2, 15–95%) | 8 | res_multistep | simple | 10 | 1.0 | Yes | Yes | No | 532.431 |
| MM-R2V-67 | O cleanup control | Clean single-reference graph + SLA 0.85 / block 32 | 8 | res_multistep | simple | 10 | 1.0 | Yes | Yes | No | 538.369 |
| MM-R2V-68 | P high resolution | 960×544 + SLA 0.85 + Comfy Kitchen backend + Low-VRAM 4 + Chunk FF 2 | 8 | res_multistep | simple | 10 | 1.0 | Yes | Yes | Yes | 659.169 |
| MM-R2V-69 | Q connected continuation | MM-R2V-68 tail context + 15-second continuation with squirrel prompt | 8 | res_multistep | simple | 15 | 1.0 | Queued | — | No | — |
## Result requirements
@@ -3,22 +3,21 @@
"2":{"inputs":{"clip_name":"qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors","type":"minimax","device":"default"},"class_type":"CLIPLoader"},
"3":{"inputs":{"vae_name":"minimax_h3_video_vae_int8_convrot.safetensors"},"class_type":"VAELoader"},
"4":{"inputs":{"vae_name":"minimax_h3_audio_vae_fp32.safetensors"},"class_type":"VAELoader"},
"5":{"inputs":{"video":"continuation_source.mp4","force_rate":24,"custom_width":0,"custom_height":0,"frame_load_cap":0,"skip_first_frames":0,"select_every_nth":1,"vae":["3",0]},"class_type":"VHS_LoadVideoPath"},
"5":{"inputs":{"video":"continuation_source.mp4","force_rate":24,"custom_width":0,"custom_height":0,"frame_load_cap":0,"skip_first_frames":0,"select_every_nth":1,"vae":["3",0]},"class_type":"VHS_LoadVideo"},
"6":{"inputs":{"image":["5",0],"batch_index":338,"length":22},"class_type":"ImageFromBatch"},
"7":{"inputs":{"image":"reference.png"},"class_type":"LoadImage"},
"8":{"inputs":{"value":"Use Picture 1 as the sole character reference and continue the preceding clip seamlessly. Preserve the subject's identity, facial features, beard, skin tone, hairstyle, white cap, white clothing, tattoos, layered chains, wrist jewelry, rings, microphone, body proportions, and overall visual design.\n\nContinue the bright sunny park scene with the subject seated on the wooden bench feeding small European robins. During this continuation, a friendly grey squirrel cautiously approaches the bench and joins the feeding scene. The subject notices the squirrel and gently offers it a small piece of food while continuing to feed the robins. Keep the squirrel anatomically natural and consistently scaled, with realistic fur, tail movement, paws, and alert but calm behavior. The robins continue natural hopping and short fluttering movements.\n\nMaintain the same warm afternoon sunlight, leafy green park, soft grass, wildflowers, blurred path, camera position, lens perspective, color palette, clothing, accessories, and cinematic realism from the preceding clip. Motion should continue smoothly from the final context frames with stable anatomy, coherent lighting, natural breathing, subtle head and hand movement, and a gentle stable camera. No scene reset, identity drift, clothing changes, extra people, duplicated animals, deformed limbs, sudden camera cut, dark lighting, or exaggerated animal behavior.","class_type":"PrimitiveStringMultiline"}},
"8":{"inputs":{"value":"Continue the preceding bright sunny park clip seamlessly. Preserve the adult male subject's identity, beard, white cap, white clothing, tattoos, layered chains, jewelry, microphone, proportions, camera, lighting, and setting. The subject remains seated on the park bench feeding red-breasted robins. During this continuation, a friendly anatomically natural grey squirrel cautiously approaches and joins the feeding scene; the subject gently offers it a small piece of food while continuing to feed the robins. Keep realistic fur, tail and paw motion, stable anatomy, natural animal behavior, coherent sunlight, smooth motion from the final context frames, and no scene reset, identity drift, extra people, duplicated animals, deformed limbs, or dark lighting."},"class_type":"PrimitiveStringMultiline"},
"9":{"inputs":{"model":["1",0],"sparsity_ratio":0.85,"block_size":"32"},"class_type":"H3SLAAttention"},
"10":{"inputs":{"model":["9",0],"attention":"comfy kitchen attention"},"class_type":"ModelAttentionBackend"},
"11":{"inputs":{"model":["10",0],"head_chunks":4},"class_type":"MiniMaxLowVRAMAttention"},
"12":{"inputs":{"model":["11",0],"chunks":2,"seq_threshold":4096},"class_type":"MiniMaxChunkFeedForward"},
"13":{"inputs":{"model":["12",0],"lora_name":"minimax\\KJminimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors","strength_model":1.0},"class_type":"LoraLoaderModelOnly"},
"14":{"inputs":{"clip":["2",0],"video_vae":["3",0],"audio_vae":["4",0],"prompt":["8",0],"chunk_seconds":15.0,"width":960,"height":544,"context_frames":"22","continuity":true,"start_frame":0,"previous_tail":["6",0],"reference_image":["7",0],"use_built_in_audio":false},"class_type":"VRGDG_MiniMaxH3ConnectedChunk"},
"15":{"inputs":{"positive":["14",0],"latent":["14",1],"context_latent":["5",4],"frame_idx":0,"include_audio":false},"class_type":"VRGDG_MiniMaxH3ApplyLatentGuide"},
"16":{"inputs":{"sampler_name":"res_multistep"},"class_type":"KSamplerSelect"},
"17":{"inputs":{"model":["13",0],"steps":8,"scheduler":"simple","denoise":1.0},"class_type":"BasicScheduler"},
"18":{"inputs":{"noise_seed":20260927},"class_type":"RandomNoise"},
"19":{"inputs":{"model":["13",0],"conditioning":["15",0]},"class_type":"BasicGuider"},
"20":{"inputs":{"noise":["18",0],"guider":["19",0],"sampler":["16",0],"sigmas":["17",0],"latent_image":["15",1]},"class_type":"SamplerCustomAdvanced"},
"19":{"inputs":{"model":["13",0],"conditioning":["14",0]},"class_type":"BasicGuider"},
"20":{"inputs":{"noise":["18",0],"guider":["19",0],"sampler":["16",0],"sigmas":["17",0],"latent_image":["14",1]},"class_type":"SamplerCustomAdvanced"},
"21":{"inputs":{"samples":["20",0],"vae":["3",0]},"class_type":"VAEDecode"},
"22":{"inputs":{"samples":["20",0],"vae":["4",0]},"class_type":"VAEDecodeAudio"},
"23":{"inputs":{"images":["21",0],"audio":["22",0],"fps":24.0,"bit_depth":8,"color_space":"sRGB","codec":"none","filename_prefix":"MiniMax/connected-squirrel-continuation"},"class_type":"CreateVideo"},
@@ -12,8 +12,10 @@ park-and-robins scene.
frames are loaded as the motion tail (`batch_index: 338`, `length: 22`).
3. If the source duration differs, change `batch_index` to
`(source_frames - 22)`.
4. The source video loader's latent output is used as the continuation
context by `VRGDG_MiniMaxH3ApplyLatentGuide`.
4. The connected-chunk node uses the visual tail directly and provides the
continuation latent to the sampler. The server's video loader does not
expose a latent output, so the optional latent-guide node is not used in
this tested template.
## Fixed generation settings