Refocus LTX 2.3 IC LoRA
Le Refocus IC LoRA pour LTX 2.3 permet de ramener des séquences légèrement floues à une netteté exploitable. Même dans des cas plus extrêmes, il peut récupérer des détails qui seraient normalement perdus, ouvrant de nouvelles possibilités pour exploiter des séquences qui auraient autrement été inutilisées.
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 previewFichiers de ce workflow
- Sortie
- Vidéo
- Auteur
- ComfyUI
Appelez 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({
"5060": {
"class_type": "VHS_LoadVideo",
"inputs": {
"custom_height": 0,
"custom_width": 0,
"force_rate": 0,
"format": "AnimateDiff",
"frame_load_cap": 0,
"select_every_nth": 1,
"skip_first_frames": 0,
"video": "UPLOADED_video_FILENAME"
},
"_meta": {
"title": "Load Video (Upload) 🎥🅥🅗🅢"
}
},
"5069": {
"class_type": "VHS_VideoCombine",
"inputs": {
"audio": [
"5060",
2
],
"crf": 13,
"filename_prefix": "outpaint_comparison",
"format": "video/h264-mp4",
"frame_rate": [
"5151:5137",
5
],
"images": [
"5151:5125",
0
],
"loop_count": 0,
"pingpong": False,
"pix_fmt": "yuv420p",
"save_metadata": True,
"save_output": True,
"trim_to_audio": False
},
"_meta": {
"title": "video_compare"
}
},
"5152": {
"class_type": "VHS_VideoCombine",
"inputs": {
"audio": [
"5060",
2
],
"crf": 13,
"filename_prefix": "outpaint_output",
"format": "video/h264-mp4",
"frame_rate": [
"5151:5137",
5
],
"images": [
"5151:5143",
0
],
"loop_count": 0,
"pingpong": False,
"pix_fmt": "yuv420p",
"save_metadata": True,
"save_output": True,
"trim_to_audio": False
},
"_meta": {
"title": "video_output"
}
},
"5151:5097": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 1920
},
"_meta": {
"title": "target_resolution_w"
}
},
"5151:5098": {
"class_type": "PrimitiveInt",
"inputs": {
"value": 1080
},
"_meta": {
"title": "target_resolution_h"
}
},
"5151:5110": {
"class_type": "LTXVConditioning",
"inputs": {
"frame_rate": [
"5151:5137",
5
],
"negative": [
"5151:5131",
0
],
"positive": [
"5151:5138",
0
]
},
"_meta": {
"title": "LTXVConditioning"
}
},
"5151:5111": {
"class_type": "LTXVConcatAVLatent",
"inputs": {
"audio_latent": [
"5151:5121",
0
],
"video_latent": [
"5151:5142",
2
]
},
"_meta": {
"title": "Concat AV Latent"
}
},
"5151:5112": {
"class_type": "EmptyLTXVLatentVideo",
"inputs": {
"batch_size": 1,
"height": [
"5151:5098",
0
],
"length": [
"5151:5126",
2
],
"width": [
"5151:5097",
0
]
},
"_meta": {
"title": "Empty LTXV Latent Video"
}
},
"5151:5113": {
"class_type": "LTXVCropGuides",
"inputs": {
"latent": [
"5151:5114",
0
],
"negative": [
"5151:5142",
1
],
"positive": [
"5151:5142",
0
]
},
"_meta": {
"title": "LTXVCropGuides"
}
},
"5151:5114": {
"class_type": "LTXVSeparateAVLatent",
"inputs": {
"av_latent": [
"5151:5122",
0
]
},
"_meta": {
"title": "Separate AV Latent"
}
},
"5151:5115": {
"class_type": "CM_FloatToInt",
"inputs": {
"a": [
"5151:5137",
5
]
},
"_meta": {
"title": "FloatToInt"
}
},
"5151:5116": {
"class_type": "CFGGuider",
"inputs": {
"cfg": 1,
"model": [
"5151:5146",
0
],
"negative": [
"5151:5142",
1
],
"positive": [
"5151:5142",
0
]
},
"_meta": {
"title": "CFG Guider"
}
},
"5151:5117": {
"class_type": "RandomNoise",
"inputs": {
"noise_seed": 42
},
"_meta": {
"title": "RandomNoise"
}
},
"5151:5118": {
"class_type": "ManualSigmas",
"inputs": {
"sigmas": "1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0"
},
"_meta": {
"title": "ManualSigmas"
}
},
"5151:5119": {
"class_type": "KSamplerSelect",
"inputs": {
"sampler_name": "euler_ancestral"
},
"_meta": {
"title": "KSamplerSelect"
}
},
"5151:5120": {
"class_type": "LTXVPreprocess",
"inputs": {
"image": [
"5151:5136",
0
],
"img_compression": 18
},
"_meta": {
"title": "LTXV Preprocess"
}
},
"5151:5121": {
"class_type": "LTXVEmptyLatentAudio",
"inputs": {
"audio_vae": [
"5151:5132",
0
],
"batch_size": 1,
"frame_rate": [
"5151:5115",
0
],
"frames_number": [
"5060",
1
]
},
"_meta": {
"title": "Empty LTXV Latent Audio"
}
},
"5151:5122": {
"class_type": "SamplerCustomAdvanced",
"inputs": {
"guider": [
"5151:5116",
0
],
"latent_image": [
"5151:5111",
0
],
"noise": [
"5151:5117",
0
],
"sampler": [
"5151:5119",
0
],
"sigmas": [
"5151:5118",
0
]
},
"_meta": {
"title": "SamplerCustomAdvanced"
}
},
"5151:5124": {
"class_type": "LTXVAudioVAEDecode",
"inputs": {
"audio_vae": [
"5151:5132",
0
],
"samples": [
"5151:5114",
1
]
},
"_meta": {
"title": "LTXV Audio VAE Decode"
}
},
"5151:5125": {
"class_type": "ImageConcanate",
"inputs": {
"direction": "left",
"image1": [
"5151:5143",
0
],
"image2": [
"5151:5130",
0
],
"match_image_size": True
},
"_meta": {
"title": "ImageConcanate"
}
},
"5151:5126": {
"class_type": "GetImageSize",
"inputs": {
"image": [
"5151:5130",
0
]
},
"_meta": {
"title": "Get Image Size"
}
},
"5151:5127": {
"class_type": "PrimitiveBoolean",
"inputs": {
"value": True
},
"_meta": {
"title": "bypass_i2v"
}
},
"5151:5128": {
"class_type": "VAEDecodeTiled",
"inputs": {
"overlap": 64,
"temporal_overlap": 4,
"temporal_size": 512,
"tile_size": 512
},
"_meta": {
"title": "VAE Decode (Tiled)"
}
},
"5151:5129": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"5151:5139",
0
],
"resize_type": "scale by multiplier",
"resize_type.multiplier": 0.5,
"scale_method": "lanczos"
},
"_meta": {
"title": "Resize Image/Mask"
}
},
"5151:5130": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"5151:5129",
0
],
"resize_type": "scale to multiple",
"resize_type.multiple": 32,
"scale_method": "area"
},
"_meta": {
"title": "Resize Image/Mask"
}
},
"5151:5131": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"5151:5135",
0
],
"text": "pc game, console game, video game, ugly, 3d render, photo, still, static, slow"
},
"_meta": {
"title": "CLIP Text Encode (Negative Prompt)"
}
},
"5151:5132": {
"class_type": "LTXVAudioVAELoader",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev-fp8.safetensors"
},
"_meta": {
"title": "Load LTXV Audio VAE"
}
},
"5151:5133": {
"class_type": "CheckpointLoaderSimple",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev-fp8.safetensors"
},
"_meta": {
"title": "Load Checkpoint"
}
},
"5151:5134": {
"class_type": "LoraLoaderModelOnly",
"inputs": {
"lora_name": "ltx-2.3-22b-distilled-lora-384.safetensors",
"model": [
"5151:5133",
0
],
"strength_model": 0.5
},
"_meta": {
"title": "Load LoRA"
}
},
"5151:5135": {
"class_type": "LTXAVTextEncoderLoader",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev-fp8.safetensors",
"device": "default",
"text_encoder": "gemma_3_12B_it.safetensors"
},
"_meta": {
"title": "Load LTXV Audio Text Encoder"
}
},
"5151:5136": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"5151:5145",
0
],
"resize_type": "match size",
"resize_type.crop": "center",
"resize_type.match": [
"5151:5130",
0
],
"scale_method": "area"
},
"_meta": {
"title": "Resize Image/Mask"
}
},
"5151:5137": {
"class_type": "VHS_VideoInfo",
"inputs": {
"video_info": [
"5060",
3
]
},
"_meta": {
"title": "Video Info 🎥🅥🅗🅢"
}
},
"5151:5138": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"5151:5135",
0
],
"text": "An aerial view gliding over a terraced stone hill town at golden hour. Clay barrel-tiled rooftops with chipped edges and lichen, weathered sandstone walls, arched windows with wooden shutters, a bell tower, narrow stepped lanes, tall cypress trees and silver-green olive orchards on the slopes. Swallows circle the roofs, laundry moves on a line and a thin curl of smoke rises from a chimney. Warm low sunlight with long crisp shadows, a lilac and apricot sky. Every surface is sharp and finely detailed: individual tiles, stones, leaves and window frames in clear focus."
},
"_meta": {
"title": "CLIP Text Encode (Positive Prompt)"
}
},
"5151:5139": {
"class_type": "ImagePadKJ",
"inputs": {
"bottom": 0,
"color": "0, 0, 0",
"extra_padding": 0,
"image": [
"5060",
0
],
"left": 0,
"pad_mode": "color",
"right": 0,
"target_height": [
"5151:5150",
1
],
"target_width": [
"5151:5149",
1
],
"top": 0
},
"_meta": {
"title": "ImagePad KJ"
}
},
"5151:5140": {
"class_type": "VHS_VideoCombine",
"inputs": {
"audio": [
"5060",
2
],
"crf": 12,
"filename_prefix": "AnimateDiff",
"format": "video/h264-mp4",
"frame_rate": [
"5151:5137",
5
],
"images": [
"5151:5144",
0
],
"loop_count": 0,
"pingpong": False,
"pix_fmt": "yuv420p",
"save_metadata": True,
"save_output": False,
"trim_to_audio": False
},
"_meta": {
"title": "Video Combine 🎥🅥🅗🅢"
}
},
"5151:5141": {
"class_type": "LTXVImgToVideoConditionOnly",
"inputs": {
"bypass": [
"5151:5127",
0
],
"image": [
"5151:5120",
0
],
"latent": [
"5151:5112",
0
],
"strength": 0.7,
"vae": [
"5151:5133",
2
]
},
"_meta": {
"title": "🅛🅣🅧 LTXV Img To Video Condition Only"
}
},
"5151:5142": {
"class_type": "LTXAddVideoICLoRAGuide",
"inputs": {
"crop": "disabled",
"frame_idx": 0,
"image": [
"5151:5144",
0
],
"latent": [
"5151:5141",
0
],
"latent_downscale_factor": 1,
"negative": [
"5151:5110",
1
],
"positive": [
"5151:5110",
0
],
"strength": 1,
"tile_frames": 97,
"tile_overlap": 64,
"tile_size": 256,
"use_streaming": False,
"use_tiled_encode": False,
"vae": [
"5151:5133",
2
]
},
"_meta": {
"title": "🅛🅣🅧 Add Video IC-LoRA Guide"
}
},
"5151:5143": {
"class_type": "Float32ColorCorrect",
"inputs": {
"brightness": 0,
"clamp_output": False,
"contrast": 1,
"exposure": 0,
"gain_b": 1,
"gain_g": 1,
"gain_r": 1,
"gamma": 0.5,
"image": [
"5151:5154",
0
],
"lift_b": 0,
"lift_g": 0,
"lift_r": 0,
"luma_space": "Rec.709 / sRGB",
"saturation": 1
},
"_meta": {
"title": "◎ Radiance Float32 Color Correct"
}
},
"5151:5144": {
"class_type": "Float32ColorCorrect",
"inputs": {
"brightness": 0,
"clamp_output": False,
"contrast": 1,
"exposure": 0,
"gain_b": 1,
"gain_g": 1,
"gain_r": 1,
"gamma": 2,
"image": [
"5151:5156",
0
],
"lift_b": 0,
"lift_g": 0,
"lift_r": 0,
"luma_space": "Rec.709 / sRGB",
"saturation": 1
},
"_meta": {
"title": "◎ Radiance Float32 Color Correct"
}
},
"5151:5145": {
"class_type": "LoadImage",
"inputs": {
"image": "642ef66b933861a25a7a8ad3b46f3a7005c783d53c3971392a84595ae4dd4b04.png"
},
"_meta": {
"title": "Load Image"
}
},
"5151:5146": {
"class_type": "LoraLoaderModelOnly",
"inputs": {
"lora_name": "ltx-2.3-22b-ic-lora-refocus.safetensors",
"model": [
"5151:5134",
0
],
"strength_model": 1
},
"_meta": {
"title": "Load LoRA"
}
},
"5151:5148": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "a / b",
"values.a": [
"5151:5097",
0
],
"values.b": [
"5151:5098",
0
]
},
"_meta": {
"title": "ratio"
}
},
"5151:5149": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "max(a, b * c)",
"values.a": [
"5151:5137",
3
],
"values.b": [
"5151:5137",
4
],
"values.c": [
"5151:5148",
0
]
},
"_meta": {
"title": "target_w"
}
},
"5151:5150": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "max(b, a / c)",
"values.a": [
"5151:5137",
3
],
"values.b": [
"5151:5137",
4
],
"values.c": [
"5151:5148",
0
]
},
"_meta": {
"title": "target_h"
}
},
"5151:5154": {
"class_type": "VAEDecode",
"inputs": {
"samples": [
"5151:5113",
2
],
"vae": [
"5151:5133",
2
]
},
"_meta": {
"title": "VAE Decode"
}
},
"5151:5156": {
"class_type": "ImageResizeKJv2",
"inputs": {
"crop_position": "center",
"device": "cpu",
"divisible_by": 2,
"height": [
"5151:5098",
0
],
"image": [
"5060",
0
],
"keep_proportion": "crop",
"pad_color": "0, 0, 0",
"upscale_method": "lanczos",
"width": [
"5151:5097",
0
]
},
"_meta": {
"title": "Resize Image v2"
}
}
})
workflow.set_input("5060", "video", client.assets.from_url("https://comfy.org/catalog-media/f1376d64-21a8-49e7-8512-2b1bc24b61f8/5d6c7957-e695-487b-b6fe-4ba20072a780"))
# 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("5152"):
output.to_file(output.name)
for output in job.get_outputs("5069"):
output.to_file(output.name)Essayer un exemple
À propos de ce workflow
Ce workflow restaure des séquences floues ou légèrement hors focus en utilisant Refocus IC LoRA sur le modèle LTX 2.3. Il ingère votre clip avec VHS_LoadVideo, traite chaque image via un processeur personnalisé LTX 2.3 + LoRA (UUID du nœud : 2e37ecc9-770a-4ac2-972c-e953901f49d3), puis réassemble les images améliorées en une vidéo avec VHS_VideoCombine. Le Refocus IC LoRA oriente le backbone LTX 2.3 vers la reconstruction des bords fins et des détails de fréquence moyenne que les filtres de netteté classiques ne peuvent pas récupérer, aidant à ramener les sujets à la clarté tout en minimisant les halos et artefacts.
Techniquement, le pipeline fonctionne image par image : VHS_LoadVideo décode la source et produit des tenseurs d’images ; le nœud personnalisé charge la base LTX 2.3 et applique les poids Refocus IC LoRA pour guider l’inférence vers des détails naturels et nets ; VHS_VideoCombine assemble les images traitées en un clip final selon le framerate et la qualité choisis. Comme il s’agit d’une amélioration pilotée par modèle et non d’un simple masque de netteté, il peut restaurer des textures subtiles et la définition des bords dans des séquences autrement inutilisables. Gardez la force de LoRA modérée pour un rendu réaliste, et itérez pour équilibrer netteté et artefacts potentiels.
Questions-réponses
Au lieu de renforcer les bords existants, LTX 2.3 avec Refocus IC LoRA utilise un modèle appris pour reconstruire des détails plausibles. Cela permet de récupérer des structures de fréquence moyenne que les masques de netteté classiques ne peuvent pas, mais des réglages trop forts peuvent introduire des artefacts — gardez le poids LoRA modéré.
Non. VHS_VideoCombine produit une vidéo à partir des images traitées uniquement. Exportez votre clip amélioré et rattachez l’audio original dans votre logiciel de montage ou via un outil de muxing séparé.
Les plans légèrement flous ou juste hors focus. Un flou extrême ou un fort flou de mouvement ne sera pas totalement récupéré — essayez de petites augmentations de la force LoRA et évaluez sur un court segment avant de traiter toute la vidéo.
La vitesse dépend de la résolution, de la durée du clip et de la VRAM GPU. Le traitement image par image avec un modèle génératif est exigeant ; testez d’abord sur une courte plage et envisagez le batching ou la réduction d’échelle pour des prévisualisations rapides.