les meilleurs modèles d'IA open source

Proposer un modèle

Nœuds ComfyUI : documentation et dépannage

Une documentation claire des nœuds ComfyUI les plus utilisés : ce que fait chaque nœud, toutes ses entrées et sorties, comment le câbler, les réglages qui comptent et les erreurs qu'il déclenche, avec la solution pour chacune. Écrite pour le ComfyUI actuel ; les packs de nœuds personnalisés sont signalés.

Cette page n'est pas encore traduite ; elle est affichée en anglais.

Tous les problèmes ComfyUI et leurs solutions

AD

Oubliez l'installation : des workflows ComfyUI préinstallés sur le cloud

BitVector exécute des workflows ComfyUI prêts à l'emploi sur ses propres GPU. Rien à installer, aucun nœud manquant, aucune case rouge. Ouvrez-le sur votre téléphone ou votre portable et générez en une minute.

Essayer BitVector

Chargeurs

Load Checkpoint

Load Checkpoint (CheckpointLoaderSimple) opens a .safetensors or .ckpt model file and hands out the three parts every workflow needs: the diffusion MODEL, the CLIP text encoder and the VAE.

Load LoRA (LoraLoader) applies a LoRA file to the MODEL and CLIP with separate strengths, so a style, character or concept can be added to any checkpoint without merging.

Load Diffusion Model

Load Diffusion Model (UNETLoader) loads a bare denoising network such as Flux, SD3.5, Wan or HunyuanVideo from models/diffusion_models, with a weight_dtype option for fp8 to save VRAM.

DualCLIPLoader loads two text encoders at once, for example clip_l plus t5xxl for Flux or clip_g plus t5xxl for SD3, and outputs one CLIP object for the prompt nodes.

Load VAE (VAELoader) loads a standalone VAE file from models/vae so a checkpoint with a missing or weak VAE decodes clean colours, and so split models like Flux and Wan get their decoder.

Unet Loader (GGUF)

Unet Loader (GGUF) loads quantised Q8, Q6, Q5 and Q4 .gguf versions of Flux, Wan, Hunyuan, Qwen-Image and other large transformers so they run on 6-12 GB GPUs; DualCLIPLoader (GGUF) does the same for T5 text encoders.

Conditionnement et prompts

CLIP Text Encode (Prompt)

CLIP Text Encode turns a text prompt into CONDITIONING using the model text encoder. One node holds the positive prompt, a second one the negative prompt.

Flux Guidance (FluxGuidance) writes the guidance value into the conditioning that Flux Dev was distilled to expect, replacing cfg; 3.5 is the default and 2 to 5 the useful range.

CLIP Set Last Layer

CLIP Set Last Layer is the ComfyUI equivalent of "clip skip": it stops the text encoder a few layers early (-2 is the common setting for anime and Pony models).

Conditioning (Combine) and (Concat)

Conditioning (Combine) merges two conditionings so the sampler is guided by both at once; Conditioning (Concat) joins the token sequences into one longer prompt. Related nodes set area, mask and timestep ranges.

Échantillonnage

KSampler runs the denoising loop: it takes the model, prompts and a latent and produces the finished latent image, controlled by seed, steps, cfg, sampler, scheduler and denoise.

KSampler (Advanced)

KSampler (Advanced) is KSampler with step-range control: add_noise, start_at_step, end_at_step and return_with_leftover_noise, used for base + refiner chains and two-model sampling.

AD

Lancez ce workflow depuis votre téléphone

Chaque workflow de cette page est préinstallé sur BitVector, modèles et nœuds personnalisés déjà en place. Choisissez-en un, tapez un prompt, c'est fait. Zéro configuration, rien à télécharger.

Essayer BitVector

Latent

VAE Decode converts the sampled LATENT into a pixel IMAGE with the VAE; VAE Decode (Tiled) does the same in tiles for very large images.

Empty Latent Image

Empty Latent Image creates the blank latent canvas (width, height, batch_size) that text-to-image sampling starts from; dimensions must be multiples of 8 and match the model family.

VAE Encode and VAE Encode (for Inpainting)

VAE Encode converts a pixel IMAGE into a LATENT for img2img; VAE Encode (for Inpainting) does the same while blanking the masked area so an inpainting model can repaint it.

Upscale Latent and Upscale Latent By

Upscale Latent resizes a latent to a new size (or by a factor) so a second KSampler can add detail at higher resolution: the classic hires fix.

EmptySD3LatentImage

EmptySD3LatentImage creates the 16-channel empty latent that Flux, SD3 and SD3.5 need; it replaces Empty Latent Image in those workflows.

Image

Save Image writes the IMAGE tensor to ComfyUI/output as a PNG, with the whole workflow embedded in the file metadata so the picture can be dragged back into ComfyUI to restore the graph.

Load Image reads a picture from the ComfyUI input folder (or an upload) and outputs an IMAGE tensor plus a MASK taken from the alpha channel, the starting point for img2img, inpainting, ControlNet and IP-Adapter graphs.

Preview Image displays an IMAGE inside the node without saving it to the output folder (a temporary file is kept until restart), ideal for checking intermediate steps like masks, ControlNet maps and hires passes.

Agrandissement

Upscale Image (using Model)

Upscale Image (using Model) runs an ESRGAN-style upscaler (4x-UltraSharp, RealESRGAN, 4x_NMKD-Siax, 4x_foolhardy_Remacri) over an IMAGE and returns it enlarged by the model's fixed factor with added detail.

Ultimate SD Upscale

ComfyUI_UltimateSDUpscale

Ultimate SD Upscale enlarges a picture with an ESRGAN model, then re-samples it tile by tile through the diffusion model with your prompt at low denoise, adding real detail at 2x-4x without running out of VRAM.

Upscale Image and Upscale Image By

Upscale Image (ImageScale) and Upscale Image By resize an IMAGE with classic interpolation (nearest, bilinear, bicubic, lanczos, area) to an exact size or by a factor, with no AI model involved.

Load Upscale Model

Load Upscale Model (UpscaleModelLoader) loads an ESRGAN-style upscaler such as 4x-UltraSharp or RealESRGAN from models/upscale_models for use with Upscale Image (using Model).

ControlNet et guidage

Apply ControlNet

Apply ControlNet attaches a guide image and a loaded ControlNet to the positive and negative CONDITIONING with strength and start/end percentages, so the sampler follows edges, depth or pose from the guide.

IPAdapter (IP-Adapter Plus)

ComfyUI_IPAdapter_plus

The IPAdapter nodes feed a reference picture into the model as an "image prompt", transferring style, composition or a face without describing them in words; IPAdapter Unified Loader plus IPAdapter Advanced are the usual pair.

Load ControlNet Model

Load ControlNet Model reads a ControlNet (or T2I-Adapter / ControlNet Union) file from models/controlnet and outputs a CONTROL_NET object for Apply ControlNet.

Masques et inpainting

Set Latent Noise Mask

Set Latent Noise Mask attaches a MASK to a LATENT so the sampler only changes the white area: inpainting and regional regeneration with any normal checkpoint.

Load Image (as Mask)

Load Image (as Mask) turns one channel of a picture (alpha, red, green or blue) into a MASK, so black-and-white mask files or painted layers can drive inpainting and compositing.

Grow Mask, Feather Mask and Invert Mask

The core mask tools expand (Grow Mask), soften (Feather Mask) and flip (Invert Mask) a MASK so inpainted regions blend into the surrounding picture without visible seams.

Vidéo et animation

Video Combine (VideoHelperSuite)

ComfyUI-VideoHelperSuite

Video Combine turns an IMAGE batch of frames into an MP4, WebM, GIF or image sequence at a chosen frame rate, with optional audio, loop count and ping-pong, the standard output node for every video workflow.

WanImageToVideo and Wan 2.x video nodes

WanImageToVideo builds the conditioning and the empty video latent for a Wan 2.1/2.2 image-to-video render from a start image, a CLIP Vision embedding and the frame count; the surrounding graph uses Load Diffusion Model, Load CLIP, Load VAE and KSampler.

AnimateDiff Loader (Evolved)

ComfyUI-AnimateDiff-Evolved

AnimateDiff Loader adds a motion module to an SD 1.5 or SDXL model so a batch of latents is rendered as coherent animation frames, the classic route to short clips from text with Stable Diffusion.

Load Video (Upload) (VideoHelperSuite)

ComfyUI-VideoHelperSuite

Load Video (Upload) decodes a video file into an IMAGE batch of frames (with frame cap, skip and every-nth options) plus the audio and frame count, the input node for video-to-video and frame-by-frame processing.

Packs de nœuds personnalisés populaires

ComfyUI Manager

ComfyUI-Manager

ComfyUI Manager is the extension that installs, updates and fixes custom node packs and models from inside the interface, resolves missing nodes in imported workflows and snapshots your setup.

FaceDetailer (Impact Pack)

ComfyUI-Impact-Pack

FaceDetailer finds faces (or hands, eyes) with a YOLO detector, crops each one, re-renders it at full resolution with your model and prompt, and pastes it back, fixing the small, mushy faces in wide shots.

Power Lora Loader (rgthree)

Power Lora Loader stacks any number of LoRAs in one node with per-row toggles and strengths, replacing chains of Load LoRA nodes; the rgthree pack also adds Fast Groups Bypasser, Context nodes and the Image Comparer.

ReActor Fast Face Swap

ComfyUI-ReActor

ReActor swaps a face from a source photo onto the generated picture with InsightFace, with optional restoration (GFPGAN, CodeFormer) and face boosting, for consistent characters without training a LoRA.

Efficient Loader and KSampler (Efficient)

efficiency-nodes-comfyui

Efficient Loader combines checkpoint, VAE, LoRA stack, both prompts and the empty latent into one node, and KSampler (Efficient) adds live previews, a built-in VAE decode and XY Plot support for comparing settings.

was-node-suite-comfyui

WAS Node Suite is a 200-node toolbox for text (concatenate, find/replace, random prompt), images (blend, filters, resize, crop, save as JPEG/WebP), masks, numbers and file handling, filling gaps the core node set leaves open.

Utilitaires et aides de workflow

ModelSamplingFlux, ModelSamplingSD3 and ModelSamplingAuraFlow

The ModelSampling nodes set the noise schedule shift for flow-matching models (Flux, SD3.5, Wan, Hunyuan, Qwen-Image, Chroma, Lumina) so samplers spend the right amount of time on composition versus detail.

Primitive, Reroute, Note and Group nodes

Primitive nodes share one value (seed, steps, prompt text, resolution) across many nodes, Reroute tidies wires, Notes document the graph and Groups let you bypass or mute whole sections at once.

AD

Marre de réparer des nœuds ? Laissez faire le cloud

BitVector garde des centaines de workflows ComfyUI installés, à jour et testés sur des GPU cloud rapides. Pas de Python, pas d'erreurs CUDA, pas de limite de VRAM. Depuis n'importe quel navigateur.

Essayer BitVector

© ModelTrends.ai

|

Fabriqué au Japon

|

© 2026