Eliminar marca de agua de video - LTX 2.3 LoRA
Sube un video con marcas de agua y elimínalas con un solo clic.
Dentro del flujo de trabajo
Una vista previa de solo lectura de la plantilla de origen. Ábrela en Cloud para explorar los nodos y hacer cambios.
Loading the graph
Open full-size workflow previewArchivos de este flujo de trabajo
- Salida
- Video
- Autor
- Rob
Llama a este modelo desde tu código
La solicitud a continuación utiliza el mismo grafo preparado y la misma configuración que el Playground. Cada envío puede iniciar una ejecución de pago; no reintentes automáticamente un envío cuyo resultado se desconozca.
What you need
- Runs on
- Comfy Cloud
- Clave
- COMFY_API_KEY
- Tus archivos
- El código los sube desde sus URL
Requiere un plan de pago de Cloud y créditos.
# 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({
"5159": {
"class_type": "SaveVideo",
"inputs": {
"filename_prefix": "video",
"format": "auto",
"format.codec": "auto",
"video": [
"5161:5062",
0
]
},
"_meta": {
"title": "Save Video"
}
},
"5160": {
"class_type": "LoadVideo",
"inputs": {
"file": "UPLOADED_video_FILENAME"
},
"_meta": {
"title": "Load Video"
}
},
"5162": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"5160",
0
]
},
"_meta": {
"title": "Get Video Components"
}
},
"5175": {
"class_type": "SaveVideo",
"inputs": {
"filename_prefix": "video_compare",
"format": "auto",
"format.codec": "auto",
"video": [
"5174:80",
0
]
},
"_meta": {
"title": "Save Comparison Video"
}
},
"5161:5057": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"5161:5084",
0
],
"text": "pc game, console game, video game, cartoon, childish, ugly"
},
"_meta": {
"title": "CLIP Text Encode (Negative Prompt)"
}
},
"5161:5058": {
"class_type": "RandomNoise",
"inputs": {
"noise_seed": 1
},
"_meta": {
"title": "RandomNoise"
}
},
"5161:5059": {
"class_type": "LatentUpscaleModelLoader",
"inputs": {
"model_name": "ltx-2.3-spatial-upscaler-x2-1.1.safetensors"
},
"_meta": {
"title": "Load Latent Upscale Model"
}
},
"5161:5060": {
"class_type": "LTXVConditioning",
"inputs": {
"frame_rate": [
"5161:5088",
2
],
"negative": [
"5161:5057",
0
],
"positive": [
"5161:5091",
0
]
},
"_meta": {
"title": "LTXVConditioning"
}
},
"5161:5061": {
"class_type": "CFGGuider",
"inputs": {
"cfg": 1,
"model": [
"5161:5087",
0
],
"negative": [
"5161:5098",
1
],
"positive": [
"5161:5098",
0
]
},
"_meta": {
"title": "CFG Guider"
}
},
"5161:5062": {
"class_type": "CreateVideo",
"inputs": {
"audio": [
"5162",
1
],
"bit_depth": "auto",
"codec": "none",
"color_space": "sRGB",
"fps": [
"5161:5088",
2
],
"images": [
"5161:5090",
0
]
},
"_meta": {
"title": "Create Video"
}
},
"5161:5063": {
"class_type": "KSamplerSelect",
"inputs": {
"sampler_name": "euler"
},
"_meta": {
"title": "KSamplerSelect"
}
},
"5161:5064": {
"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"
}
},
"5161:5065": {
"class_type": "LTXVEmptyLatentAudio",
"inputs": {
"audio_vae": [
"5161:5086",
0
],
"batch_size": 1,
"frame_rate": [
"5161:5095",
0
],
"frames_number": [
"5161:5097",
1
]
},
"_meta": {
"title": "Empty LTXV Latent Audio"
}
},
"5161:5066": {
"class_type": "LTXVConcatAVLatent",
"inputs": {
"audio_latent": [
"5161:5065",
0
],
"video_latent": [
"5161:5094",
2
]
},
"_meta": {
"title": "Concat AV Latent"
}
},
"5161:5067": {
"class_type": "SamplerCustomAdvanced",
"inputs": {
"guider": [
"5161:5061",
0
],
"latent_image": [
"5161:5066",
0
],
"noise": [
"5161:5058",
0
],
"sampler": [
"5161:5063",
0
],
"sigmas": [
"5161:5064",
0
]
},
"_meta": {
"title": "SamplerCustomAdvanced"
}
},
"5161:5068": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"5161:5088",
0
],
"resize_type": "scale dimensions",
"resize_type.crop": "disabled",
"resize_type.height": 1088,
"resize_type.width": 1920,
"scale_method": "area"
},
"_meta": {
"title": "Resize Image/Mask (s2 size)"
}
},
"5161:5069": {
"class_type": "LTXVLatentUpsampler",
"inputs": {
"samples": [
"5161:5082",
2
],
"upscale_model": [
"5161:5059",
0
],
"vae": [
"5161:5085",
2
]
},
"_meta": {
"title": "LTXVLatentUpsampler"
}
},
"5161:5070": {
"class_type": "LTXVSeparateAVLatent",
"inputs": {
"av_latent": [
"5161:5067",
0
]
},
"_meta": {
"title": "Separate AV Latent"
}
},
"5161:5071": {
"class_type": "LTXAddVideoICLoRAGuide",
"inputs": {
"crop": "disabled",
"frame_idx": 0,
"image": [
"5161:5068",
0
],
"latent": [
"5161:5069",
0
],
"latent_downscale_factor": 1,
"negative": [
"5161:5099",
1
],
"positive": [
"5161:5099",
0
],
"strength": 1,
"tile_frames": 97,
"tile_overlap": 64,
"tile_size": 256,
"use_streaming": False,
"use_tiled_encode": True,
"vae": [
"5161:5085",
2
]
},
"_meta": {
"title": "🅛🅣🅧 Add Video IC-LoRA Guide"
}
},
"5161:5072": {
"class_type": "KSamplerSelect",
"inputs": {
"sampler_name": "euler"
},
"_meta": {
"title": "KSamplerSelect"
}
},
"5161:5073": {
"class_type": "SamplerCustomAdvanced",
"inputs": {
"guider": [
"5161:5076",
0
],
"latent_image": [
"5161:5077",
0
],
"noise": [
"5161:5078",
0
],
"sampler": [
"5161:5072",
0
],
"sigmas": [
"5161:5075",
0
]
},
"_meta": {
"title": "SamplerCustomAdvanced"
}
},
"5161:5074": {
"class_type": "LTXVCropGuides",
"inputs": {
"latent": [
"5161:5080",
0
],
"negative": [
"5161:5071",
1
],
"positive": [
"5161:5071",
0
]
},
"_meta": {
"title": "LTXVCropGuides"
}
},
"5161:5075": {
"class_type": "ManualSigmas",
"inputs": {
"sigmas": "0.909375, 0.725, 0.421875, 0.0"
},
"_meta": {
"title": "ManualSigmas"
}
},
"5161:5076": {
"class_type": "CFGGuider",
"inputs": {
"cfg": 1,
"model": [
"5161:5087",
0
],
"negative": [
"5161:5071",
1
],
"positive": [
"5161:5071",
0
]
},
"_meta": {
"title": "CFG Guider"
}
},
"5161:5077": {
"class_type": "LTXVConcatAVLatent",
"inputs": {
"audio_latent": [
"5161:5099",
2
],
"video_latent": [
"5161:5071",
2
]
},
"_meta": {
"title": "Concat AV Latent"
}
},
"5161:5078": {
"class_type": "RandomNoise",
"inputs": {
"noise_seed": 29
},
"_meta": {
"title": "RandomNoise"
}
},
"5161:5079": {
"class_type": "LTXVAudioVAEDecode",
"inputs": {
"audio_vae": [
"5161:5086",
0
],
"samples": [
"5161:5080",
1
]
},
"_meta": {
"title": "LTXV Audio VAE Decode"
}
},
"5161:5080": {
"class_type": "LTXVSeparateAVLatent",
"inputs": {
"av_latent": [
"5161:5073",
0
]
},
"_meta": {
"title": "Separate AV Latent"
}
},
"5161:5082": {
"class_type": "LTXVCropGuides",
"inputs": {
"latent": [
"5161:5070",
0
],
"negative": [
"5161:5094",
1
],
"positive": [
"5161:5094",
0
]
},
"_meta": {
"title": "LTXVCropGuides"
}
},
"5161:5083": {
"class_type": "LTXVAudioVAEEncode",
"inputs": {
"audio": [
"5161:5088",
1
],
"audio_vae": [
"5161:5086",
0
]
},
"_meta": {
"title": "LTXV Audio VAE Encode"
}
},
"5161:5084": {
"class_type": "LTXAVTextEncoderLoader",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev.safetensors",
"device": "default",
"text_encoder": "gemma_3_12B_it.safetensors"
},
"_meta": {
"title": "Load LTXV Audio Text Encoder"
}
},
"5161:5085": {
"class_type": "CheckpointLoaderSimple",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev.safetensors"
},
"_meta": {
"title": "Load Checkpoint"
}
},
"5161:5086": {
"class_type": "LTXVAudioVAELoader",
"inputs": {
"ckpt_name": "ltx-2.3-22b-dev.safetensors"
},
"_meta": {
"title": "Load LTXV Audio VAE"
}
},
"5161:5087": {
"class_type": "LTXICLoRALoaderModelOnly",
"inputs": {
"lora_name": "ltx2.3-ic-watermark-remove-general.safetensors",
"model": [
"5161:5085",
0
],
"strength_model": 1.5
},
"_meta": {
"title": "🅛🅣🅧 IC-LoRA Loader Model Only"
}
},
"5161:5088": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"5160",
0
]
},
"_meta": {
"title": "Get Video Components"
}
},
"5161:5089": {
"class_type": "ResizeImageMaskNode",
"inputs": {
"input": [
"5161:5088",
0
],
"resize_type": "scale dimensions",
"resize_type.crop": "disabled",
"resize_type.height": 544,
"resize_type.width": 960,
"scale_method": "area"
},
"_meta": {
"title": "Resize Image/Mask (s1 size)"
}
},
"5161:5090": {
"class_type": "LTXVTiledVAEDecode",
"inputs": {
"horizontal_tiles": 2,
"last_frame_fix": False,
"latents": [
"5161:5074",
2
],
"overlap": 6,
"vae": [
"5161:5085",
2
],
"vertical_tiles": 2,
"working_device": "auto",
"working_dtype": "auto"
},
"_meta": {
"title": "🅛🅣🅧 LTXV Tiled VAE Decode"
}
},
"5161:5091": {
"class_type": "CLIPTextEncode",
"inputs": {
"clip": [
"5161:5084",
0
],
"text": "Remove the short-video platform watermarks and every related occlusion from this clip: the semi-transparent account handle that drifts across the frame and jumps to a new position halfway through, and the small rounded logo badge with its name in the top-left corner. Restore what was hidden underneath them so the picture looks untouched: the wet cobblestones, the ornate green streetlamp, the limestone facades with their shutters and balconies, the swirling golden leaves and the dancing man in his mustard trousers and striped sweater. Keep the original motion, framing, colour and golden-hour light exactly as they are, with clean, sharp, natural detail and no blur, smudges or ghosting where the overlays used to be."
},
"_meta": {
"title": "CLIP Text Encode (Positive Prompt)"
}
},
"5161:5093": {
"class_type": "EmptyLTXVLatentVideo",
"inputs": {
"batch_size": 1,
"height": 544,
"length": [
"5161:5097",
1
],
"width": 960
},
"_meta": {
"title": "Empty LTXV Latent Video"
}
},
"5161:5094": {
"class_type": "LTXAddVideoICLoRAGuide",
"inputs": {
"crop": "disabled",
"frame_idx": 0,
"image": [
"5161:5089",
0
],
"latent": [
"5161:5093",
0
],
"latent_downscale_factor": 1,
"negative": [
"5161:5060",
1
],
"positive": [
"5161:5060",
0
],
"strength": 1,
"tile_frames": 97,
"tile_overlap": 64,
"tile_size": 256,
"use_streaming": False,
"use_tiled_encode": True,
"vae": [
"5161:5085",
2
]
},
"_meta": {
"title": "🅛🅣🅧 Add Video IC-LoRA Guide"
}
},
"5161:5095": {
"class_type": "LTXFloatToInt",
"inputs": {
"a": [
"5161:5088",
2
]
},
"_meta": {
"title": "🅛🅣🅧 Float To Int"
}
},
"5161:5096": {
"class_type": "GetImageSizeAndCount",
"inputs": {
"image": [
"5161:5089",
0
]
},
"_meta": {
"title": "Get Image Size & Count"
}
},
"5161:5097": {
"class_type": "ComfyMathExpression",
"inputs": {
"expression": "(int((a - 1) / 8)) * 8 + 1",
"values.a": [
"5161:5096",
3
]
},
"_meta": {
"title": "Math Expression"
}
},
"5161:5098": {
"class_type": "LTXVSetAudioRefTokens",
"inputs": {
"audio_latent": [
"5161:5083",
0
],
"negative": [
"5161:5094",
1
],
"positive": [
"5161:5094",
0
]
},
"_meta": {
"title": "🅛🅣🅧 Set Audio Ref Tokens"
}
},
"5161:5099": {
"class_type": "LTXVSetAudioRefTokens",
"inputs": {
"audio_latent": [
"5161:5070",
1
],
"negative": [
"5161:5082",
1
],
"positive": [
"5161:5082",
0
]
},
"_meta": {
"title": "🅛🅣🅧 Set Audio Ref Tokens"
}
},
"5174:77": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"5160",
0
]
},
"_meta": {
"title": "Get Video Components"
}
},
"5174:78": {
"class_type": "GetVideoComponents",
"inputs": {
"video": [
"5161:5062",
0
]
},
"_meta": {
"title": "Get Video Components"
}
},
"5174:79": {
"class_type": "ImageStitch",
"inputs": {
"direction": "down",
"image1": [
"5174:77",
0
],
"image2": [
"5174:78",
0
],
"match_image_size": False,
"spacing_color": "white",
"spacing_width": 0
},
"_meta": {
"title": "Stitch Images"
}
},
"5174:80": {
"class_type": "CreateVideo",
"inputs": {
"audio": [
"5174:77",
1
],
"bit_depth": "auto",
"codec": "none",
"color_space": "sRGB",
"fps": [
"5174:77",
2
],
"images": [
"5174:79",
0
]
},
"_meta": {
"title": "Create Video"
}
}
})
workflow.set_input("5160", "file", client.assets.from_url("https://comfy.org/catalog-media/2f731481-9bd0-4266-8d74-16ae4ccceded/b328bcd8-cf71-44aa-8ba2-b6581db74676"))
# 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("5159"):
output.to_file(output.name)
for output in job.get_outputs("5175"):
output.to_file(output.name)Probar un ejemplo
Acerca de este flujo de trabajo
Este flujo de trabajo de ComfyUI elimina marcas de agua persistentes de un video mediante inpainting cuadro por cuadro con una pasada de difusión mejorada por LoRA. Comienza con LoadVideo para cargar tu clip y GetVideoComponents para extraer los cuadros, dimensiones e información de tiempo. El procesamiento principal lo realizan dos nodos personalizados (UUIDs 3b019976-c705-40af-8f45-081975d72105 y 83cfc570-fff1-4869-8c43-d39ef189634b), que aplican un paso de inpainting guiado por LTX 2.3 LoRA a lo largo de la secuencia para reconstruir el fondo limpio donde está la marca de agua. Finalmente, SaveVideo recompone los cuadros procesados en un nuevo archivo de video.
Técnicamente, la canalización realiza inpainting de video a video usando un modelo estilo Stable Diffusion aumentado con el adaptador LTX 2.3 LoRA para orientar la reconstrucción hacia contenido plausible y suprimir la marca de agua. La etapa personalizada mantiene la consistencia por cuadro y utiliza ediciones guiadas por máscara, de modo que solo se regenere el área marcada, preservando el resto del cuadro. Al trabajar a nivel de cuadro y estar guiado por inpainting, el resultado evita los típicos desenfoques o manchas de los filtros tradicionales y funciona mejor en logotipos semitransparentes o en esquinas. Es una solución práctica y de un solo clic para limpiar material propio o licenciado sin rotoscopía manual.
Preguntas y respuestas
Utiliza una canalización de inpainting guiada por LTX 2.3 LoRA. Coloca LTX 2.3 LoRA en models/loras y asegúrate de que haya un checkpoint base compatible en models/checkpoints. El nodo personalizado (UUIDs 3b019976-c705-40af-8f45-081975d72105 y 83cfc570-fff1-4869-8c43-d39ef189634b) se encarga de aplicar LoRA a los cuadros del video.
Está optimizado para marcas de agua persistentes y estáticas (por ejemplo, logotipos en esquinas o fechas). Las superposiciones en movimiento, animadas o de gran tamaño son más difíciles de inpaintar de forma consistente y pueden requerir ajuste de parámetros, pre-mascarado o limpieza manual.
SaveVideo normalmente guarda solo el video procesado. Si necesitas el audio original, remuxéalo después usando un editor de video o un comando como: ffmpeg -i clean.mp4 -i source.mp4 -map 0:v -map 1:a -c copy final.mp4.
Prueba aumentando moderadamente el peso de LoRA y la fuerza de denoise, reduce la escala ligeramente para disminuir el ruido de procesamiento y asegúrate de que la región de la marca de agua esté completamente cubierta por la máscara de inpainting en el nodo personalizado. Para casos muy difíciles, procesa en segmentos alrededor de los cortes de escena.