LTX Cleanplate pour VFX
Décrivez un objet à retirer, puis SAM3 et LTX 2.3 (inpainting conditionné par masque) l’effacent et reconstruisent le fond propre.
Dans le workflow
Un aperçu en lecture seule du modèle source. Ouvrez-le dans Cloud pour explorer les nœuds et apporter des modifications.
Loading the graph
Open full-size workflow previewAppelez ce modèle depuis votre code
La requête ci-dessous utilise le même graphe préparé et les mêmes paramètres que le Playground. Chaque envoi peut lancer une exécution payante ; ne relancez pas automatiquement un envoi dont l’issue est inconnue.
What you need
- Runs on
- Comfy Cloud
- Clé
- COMFY_API_KEY
- Vos fichiers
- Importés par le code depuis leurs URL
Nécessite un forfait Cloud payant et des crédits.
# Python 3.10+: pip install comfy-sdk==0.4.0
import os
from comfy_sdk import Comfy
api_key = os.environ.get("COMFY_API_KEY", "").strip()
if not api_key: raise RuntimeError("Set COMFY_API_KEY")
os.environ.setdefault("COMFY_BASE_URL", "https://cloud.comfy.org")
client = Comfy(api_key=api_key)
workflow = client.workflows.from_json({
"52": {
"class_type": "SaveVideo",
"inputs": {
"codec": "auto",
"filename_prefix": "cleanplate_compare",
"format": "auto",
"format.codec": "auto",
"video": [
"54:31",
0
]
},
"_meta": {
"title": "Save Video Compared"
}
},
"53": {
"class_type": "SaveVideo",
"inputs": {
"codec": "auto",
"filename_prefix": "cleanplate_ltx_inpaint",
"format": "auto",
"format.codec": "auto",
"video": [
"54:36",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"58": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"117",
0
]
},
"_meta": {
"title": "Plate frames"
}
},
"59": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"54:36",
0
]
},
"_meta": {
"title": "LTX inpaint frames"
}
},
"63": {
"class_type": "VHS_GetImageCount",
"inputs": {
"images": [
"58",
0
]
},
"_meta": {
"title": "n"
}
},
"64": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "(int((a - 1) / 8)) * 8 + 1",
"values.a": [
"63",
0
]
},
"_meta": {
"title": "L = valid LTX length"
}
},
"65": {
"class_type": "GetImageRangeFromBatch",
"inputs": {
"images": [
"58",
0
],
"num_frames": [
"64",
1
],
"start_index": 0
},
"_meta": {
"title": "Trim plate to L"
}
},
"66": {
"class_type": "LTXVDilateVideoMask",
"inputs": {
"mask": [
"73",
0
],
"spatial_radius": 10,
"temporal_radius": 1
},
"_meta": {
"title": "Dilate matte"
}
},
"67": {
"class_type": "GrowMaskWithBlur",
"inputs": {
"blur_radius": 4,
"decay_factor": 1,
"expand": 0,
"fill_holes": False,
"flip_input": False,
"incremental_expandrate": 0,
"lerp_alpha": 1,
"mask": [
"66",
0
],
"tapered_corners": True
},
"_meta": {
"title": "Feather"
}
},
"68": {
"class_type": "LTXVInpaintPreprocess",
"inputs": {
"images": [
"65",
0
],
"mask": [
"67",
0
]
},
"_meta": {
"title": "Green-fill masked region"
}
},
"69": {
"class_type": "CreateVideo",
"inputs": {
"bit_depth": 8,
"codec": "none",
"color_space": "sRGB",
"fps": [
"58",
2
],
"images": [
"68",
0
]
},
"_meta": {
"title": "-> subgraph guide"
}
},
"70": {
"class_type": "LoadSAM3Model",
"inputs": {
"model_path": "models/sam3/sam3.pt"
},
"_meta": {
"title": "(down)Load SAM3 Model"
}
},
"71": {
"class_type": "SAM3VideoSegmentation",
"inputs": {
"frame_idx": 0,
"prompt_mode": "text",
"score_threshold": 0.3,
"text_prompt": [
"130",
0
],
"video_frames": [
"65",
0
]
},
"_meta": {
"title": "SAM3 (object prompt)"
}
},
"72": {
"class_type": "SAM3Propagate",
"inputs": {
"direction": "both",
"end_frame": -1,
"offload_model": False,
"sam3_model": [
"70",
0
],
"start_frame": 0,
"video_state": [
"71",
0
]
},
"_meta": {
"title": "SAM3 Propagate"
}
},
"73": {
"class_type": "SAM3VideoOutput",
"inputs": {
"masks": [
"72",
0
],
"obj_id": -1,
"plot_all_masks": True,
"scores": [
"72",
1
],
"video_state": [
"72",
2
]
},
"_meta": {
"title": "SAM3 Video Output"
}
},
"75": {
"class_type": "MaskToImage",
"inputs": {
"mask": [
"67",
0
]
},
"_meta": {
"title": "Convert Mask to Image"
}
},
"76": {
"class_type": "CreateVideo",
"inputs": {
"bit_depth": 8,
"codec": "none",
"color_space": "sRGB",
"fps": [
"58",
2
],
"images": [
"75",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"77": {
"class_type": "SaveVideo",
"inputs": {
"codec": "auto",
"filename_prefix": "cleanplate_matte",
"format": "auto",
"format.codec": "auto",
"video": [
"76",
0
]
},
"_meta": {
"title": "Save Matte (Nuke)"
}
},
"116": {
"class_type": "VHS_LoadVideo",
"inputs": {
"custom_height": 0,
"custom_width": 0,
"force_rate": 0,
"format": "LTXV",
"frame_load_cap": 0,
"select_every_nth": 1,
"skip_first_frames": 0,
"video": "UPLOADED_video_FILENAME"
},
"_meta": {
"title": "Load Video (Upload) 🎥🅥🅗🅢"
}
},
"117": {
"class_type": "CreateVideo",
"inputs": {
"audio": [
"116",
2
],
"bit_depth": 8,
"codec": "none",
"color_space": "sRGB",
"fps": [
"118",
0
],
"images": [
"116",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"118": {
"class_type": "VHS_VideoInfo",
"inputs": {
"video_info": [
"116",
3
]
},
"_meta": {
"title": "Video Info 🎥🅥🅗🅢"
}
},
"130": {
"class_type": "PrimitiveString",
"inputs": {
"value": "cart"
},
"_meta": {
"title": "What do you want to remove?"
}
},
"133": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"134",
0
]
},
"_meta": {
"title": "Final prompt"
}
},
"134": {
"class_type": "StringConcatenate",
"inputs": {
"delimiter": " ",
"string_a": [
"135",
0
],
"string_b": "removed"
},
"_meta": {
"title": "+ suffix"
}
},
"135": {
"class_type": "StringConcatenate",
"inputs": {
"delimiter": " ",
"string_a": "background plate, the scene with the",
"string_b": [
"130",
0
]
},
"_meta": {
"title": "prefix + object"
}
},
"54:1": {
"class_type": "VAELoaderKJ",
"inputs": {
"device": "main_device",
"vae_name": "LTX23_video_vae_bf16.safetensors",
"weight_dtype": "bf16"
},
"_meta": {
"title": "VAELoader KJ"
}
},
"54:10": {
"class_type": "EmptyLTXVLatentVideo",
"inputs": {
"batch_size": 1,
"height": [
"54:6",
1
],
"length": [
"54:45",
1
],
"width": [
"54:6",
0
]
},
"_meta": {
"title": "Empty LTXV Latent Video"
}
},
"54:11": {
"class_type": "LTXVConditioning",
"inputs": {
"frame_rate": [
"54:49",
0
],
"negative": [
"54:22",
0
],
"positive": [
"54:9",
0
]
},
"_meta": {
"title": "LTXVConditioning"
}
},
"54:12": {
"class_type": "LTXAddVideoICLoRAGuide",
"inputs": {
"crop": "disabled",
"frame_idx": 0,
"image": [
"54:41",
0
],
"latent": [
"54:10",
0
],
"latent_downscale_factor": 1,
"negative": [
"54:11",
1
],
"positive": [
"54:11",
0
],
"strength": 1,
"tile_frames": 97,
"tile_overlap": 64,
"tile_size": 256,
"use_streaming": False,
"use_tiled_encode": False,
"vae": [
"54:1",
0
]
},
"_meta": {
"title": "🅛🅣🅧 Add Video IC-LoRA Guide"
}
},
"54:126": {
"class_type": "ComfyNumberConvert",
"inputs": {
"value": [
"58",
2
]
},
"_meta": {
"title": "Convert Number"
}
},
"54:13": {
"class_type": "LTXVAudioVAEDecode",
"inputs": {
"audio_vae": [
"54:2",
0
],
"samples": [
"54:18",
1
]
},
"_meta": {
"title": "LTXV Audio VAE Decode"
}
},
"54:14": {
"class_type": "LoraLoaderModelOnly",
"inputs": {
"lora_name": "ltx-2.3-22b-distilled-lora-384-1.1.safetensors",
"model": [
"54:15",
0
],
"strength_model": 0.4
},
"_meta": {
"title": "Load LoRA"
}
},
"54:15": {
"class_type": "LoraLoaderModelOnly",
"inputs": {
"lora_name": "ltx-2.3-22b-ic-lora-in-outpainting-0.9.safetensors",
"model": [
"54:5",
0
],
"strength_model": 1
},
"_meta": {
"title": "Load LoRA"
}
},
"54:16": {
"class_type": "VAEDecode",
"inputs": {
"samples": [
"54:17",
2
],
"vae": [
"54:1",
0
]
},
"_meta": {
"title": "VAE Decode"
}
},
"54:17": {
"class_type": "LTXVCropGuides",
"inputs": {
"latent": [
"54:18",
0
],
"negative": [
"54:12",
1
],
"positive": [
"54:12",
0
]
},
"_meta": {
"title": "LTXVCropGuides"
}
},
"54:18": {
"class_type": "LTXVSeparateAVLatent",
"inputs": {
"av_latent": [
"54:19",
0
]
},
"_meta": {
"title": "Separate AV Latent"
}
},
"54:19": {
"class_type": "KSampler",
"inputs": {
"cfg": 1,
"denoise": 1,
"latent_image": [
"54:20",
0
],
"model": [
"54:14",
0
],
"negative": [
"54:12",
1
],
"positive": [
"54:12",
0
],
"sampler_name": "euler_ancestral_cfg_pp",
"scheduler": "linear_quadratic",
"seed": 511332212712579,
"steps": 8
},
"_meta": {
"title": "KSampler"
}
},
"54:2": {
"class_type": "VAELoaderKJ",
"inputs": {
"device": "main_device",
"vae_name": "LTX23_audio_vae_bf16.safetensors",
"weight_dtype": "bf16"
},
"_meta": {
"title": "VAELoader KJ"
}
},
"54:20": {
"class_type": "LTXVConcatAVLatent",
"inputs": {
"audio_latent": [
"54:21",
0
],
"video_latent": [
"54:12",
2
]
},
"_meta": {
"title": "Concat AV Latent"
}
},
"54:21": {
"class_type": "LTXVEmptyLatentAudio",
"inputs": {
"audio_vae": [
"54:2",
0
],
"batch_size": 1,
"frame_rate": [
"54:50",
0
],
"frames_number": [
"54:45",
1
]
},
"_meta": {
"title": "Empty LTXV Latent Audio"
}
},
"54:22": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"54:4",
0
],
"text": "blurry, low resolution, ugly, garbled mess, compression artifacts, glitch, stop motion, still image, pixelated, robotic voice, oversaturated"
},
"_meta": {
"title": "CLIP Text Encode (Prompt)"
}
},
"54:25": {
"class_type": "ImageFromBatch",
"inputs": {
"batch_index": 0,
"image": [
"54:41",
0
],
"length": 1
},
"_meta": {
"title": "Get Image from Batch"
}
},
"54:26": {
"class_type": "ImageStitch",
"inputs": {
"direction": "down",
"image1": [
"54:41",
0
],
"image2": [
"54:16",
0
],
"match_image_size": True,
"spacing_color": "white",
"spacing_width": 0
},
"_meta": {
"title": "Stitch Images"
}
},
"54:3": {
"class_type": "VHS_GetImageCount",
"inputs": {
"images": [
"54:38",
0
]
},
"_meta": {
"title": "Get Image Count 🎥🅥🅗🅢"
}
},
"54:31": {
"class_type": "CreateVideo",
"inputs": {
"bit_depth": 8,
"codec": "none",
"color_space": "sRGB",
"fps": [
"54:49",
0
],
"images": [
"54:26",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"54:36": {
"class_type": "CreateVideo",
"inputs": {
"audio": [
"54:38",
1
],
"bit_depth": 8,
"codec": "none",
"color_space": "sRGB",
"fps": [
"54:49",
0
],
"images": [
"54:16",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"54:38": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"69",
0
]
},
"_meta": {
"title": "Get Video Components"
}
},
"54:4": {
"class_type": "DualCLIPLoader",
"inputs": {
"clip_name1": "gemma_3_12B_it_fp4_mixed.safetensors",
"clip_name2": "ltx-2.3_text_projection_bf16.safetensors",
"device": "default",
"type": "ltxv"
},
"_meta": {
"title": "Load CLIP (Dual)"
}
},
"54:41": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"54:38",
0
],
"resize_type": "scale dimensions",
"resize_type.crop": "disabled",
"resize_type.height": [
"118",
4
],
"resize_type.width": [
"118",
3
],
"scale_method": "area"
},
"_meta": {
"title": "Resize Image/Mask (s1 size)"
}
},
"54:43": {
"class_type": "PreviewAny",
"inputs": {
"source": [
"54:45",
1
]
},
"_meta": {
"title": "Preview as Text"
}
},
"54:45": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "(int((a - 1) / 8)) * 8 + 1",
"values.a": [
"54:3",
0
]
},
"_meta": {
"title": "Math Expression"
}
},
"54:49": {
"class_type": "ComfyNumberConvert",
"inputs": {
"value": [
"54:50",
0
]
},
"_meta": {
"title": "Convert Number"
}
},
"54:5": {
"class_type": "UNETLoader",
"inputs": {
"unet_name": "ltx-2.3-22b-dev_transformer_only_bf16.safetensors",
"weight_dtype": "default"
},
"_meta": {
"title": "Load Diffusion Model"
}
},
"54:50": {
"class_type": "PrimitiveInt",
"inputs": {
"value": [
"54:126",
1
]
},
"_meta": {
"title": "fps"
}
},
"54:6": {
"class_type": "GetImageSize",
"inputs": {
"image": [
"54:25",
0
]
},
"_meta": {
"title": "Get Image Size"
}
},
"54:9": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"54:4",
0
],
"text": [
"133",
0
]
},
"_meta": {
"title": "CLIP Text Encode (Prompt)"
}
}
})
workflow.set_input("116", "video", client.assets.from_url("https://comfy.org/catalog-media/9c45ce1c-1805-4be7-ac67-11f7bb6d02fa/40969461-d6a9-41e2-9e20-eb8d27b27314"))
# run() also passes api_key to the partner nodes in this workflow.
job = client.run(workflow, api_key=api_key)
for output in job.get_outputs("53"):
output.to_file(output.name)
for output in job.get_outputs("52"):
output.to_file(output.name)
for output in job.get_outputs("77"):
output.to_file(output.name)Essayer un exemple
À propos de ce workflow
Ce workflow ComfyUI construit un fond propre (clean plate) en supprimant un objet d’une vidéo grâce à des masques générés par SAM3 et à l’inpainting conditionné par masque LTX-2.3 In‑Outpainting IC‑LoRA. Le pipeline charge votre clip avec VHS_LoadVideo/VHS_VideoInfo, extrait les images via GetVideoComponents, puis utilise LoadSAM3Model → SAM3VideoSegmentation pour générer un masque de l’objet à retirer. SAM3Propagate suit ce masque sur toute la séquence, et SAM3VideoOutput fournit un OBJECT_MASK image par image. Pour garantir un remplissage homogène, LTXVDilateVideoMask agrandit le masque (ex : s10/t1) et GrowMaskWithBlur adoucit les bords.
La zone masquée est ensuite remplie en vert par LTXVInpaintPreprocess, puis CreateVideo ré-encode ces images prétraitées. Ce clean plate vert, avec l’OBJECT_MASK et votre prompt décrivant uniquement le fond, alimente le sous-graphe LTX 2.3 In‑Outpainting IC‑LoRA, qui génère uniquement à l’intérieur du masque. Des utilitaires comme VHS_GetImageCount, ComfyMathExpression et GetImageRangeFromBatch permettent de prévisualiser une plage d’images avant un traitement complet. Enfin, SaveVideo enregistre le clean plate pour le compositing en aval.
Questions-réponses
Laissez-le vide ou décrivez uniquement le fond à l’intérieur de la zone masquée (ex : « route en asphalte avec lignes peintes, ombre douce »). N’évoquez pas l’objet supprimé ni d’éléments de premier plan non concernés — cela peut provoquer des duplications ou hallucinations.
Augmentez la taille de LTXVDilateVideoMask pour couvrir le flou de mouvement et dépasser l’objet, puis augmentez le rayon de flou de GrowMaskWithBlur pour adoucir la transition. Re-sélectionnez aussi le masque SAM3 sur une image plus nette et relancez la propagation si le suivi dérive.
Oui. Utilisez VHS_GetImageCount pour connaître le nombre total d’images et GetImageRangeFromBatch (éventuellement avec ComfyMathExpression) pour sélectionner une petite plage d’images, puis repassez sur toute la séquence une fois satisfait.
SaveVideo produit généralement un fichier muet. Pour ajouter l’audio, remuxez la piste originale dans votre logiciel de montage ou avec ffmpeg. Si votre configuration gère l’audio via GetVideoComponents, vous pouvez aussi le traiter séparément après le rendu.