ComfyUI Assistant Node
A custom ComfyUI node that integrates with the OpenAI Assistants API.
Nodes (84)
Turn a bool, an int, and a float into one tidy string
Images, latents, numbers, lists
A two-input AND/OR gate for your workflow, that's the whole node
The node this whole pack is named after
One wildcard input, a passthrough and a has_value flag
Steal the color grade from one image and slap it on another
A very specific ComfyDeploy deployment, driven from your graph
Queue an avatar-generation deployment and pull the result back
Name your own inputs
The two-second comparison node that unblocks whole workflow branches
The one node that does exactly one thing, and does it locally
Crop a subject out of a mask's bounding box, padding included
March the alpha outward
A depth map from the cloud, zero preprocessor setup
Describe an object, get its mask back — no local GPU needed
Five images, five text prompts, one parallel segmentation call
Orbit a camera around a single image with Flux.2
Flux 2 Dev in ComfyUI without the 24 GB of VRAM
The faster, cheaper way to use Flux 2 without local VRAM
The flagship, with none of the local install
The 'just generate' node
Mask-based Flux inpainting in the cloud, the honest way
One photo in, stylized avatars out — PuLID identity, zero local setup
NAG guidance, shifts, and reference images
Inpaint the masked region with Flux Pro Fill, all in the cloud
Flux Pro 1.1 Ultra with parallel requests and per-item retries built in
Put a face on a Flux image without installing InsightFace
Run your trained Flux LoRAs on fal — no local model, no VRAM squeeze
Flux, a face, and a LoRA — all from the same cloud call
Change the prompt, keep the subject
Kontext editing, masked and LoRA-loaded, no local weights
Kontext editing with your LoRA on top — style and subject in one edit
Merge two or more images with Kontext Max — one instruction, up to four references
The flagship editor, wrapped
Kontext Pro, the API-only editor, from inside ComfyUI
The fast, cheap reframe variant when you're iterating on compositions
Repaint your photo into a new aspect ratio with Luma Photon
Tell it what to find, get a mask back — no GPU, no model download
Nano Banana editing without renting a 24GB card
Qwen-Image's 20B, rented by the click
BiRefNet cutouts on demand, with the hard cases in mind
Name an object, get a SAM-3 mask — the heavyweight, hosted
ByteDance's SeeDream 4.5, without the ByteDance account
SeeDream 4 image editing, via fal and inside your graph
Batch prompts, retries, and one text output
The Gemini node built for avatar batch work — with an OpenAI escape hatch
Gemini, six images at a time — the multimodal variant
Ask Gemini 2.5 for a mask, not a caption
Gemini first, GPT as the safety net — a text node that survives outages
Six images, two providers, one text output
The node that answers 'what size is my image, actually?'
Nano Banana straight from Google — with a fal.ai parachute
Nano Banana for edits, where the image isn't optional
Nano Banana with up to eight reference images
The flagship, multi-image, with Google Search even
Un-batch your images so list-expecting nodes stop fighting you
Paste one image onto another — batch-smart, mask-aware
Turn a pile of images into one batch (and back again)
The resize node that respects your latent's multiples
Stitch up to eight images into one canvas, grid or guillotine
Measure how far a frame drifted, down to sub-pixels
Do real math inside your graph, not on a calculator
Math that picks its own branch
A pass-through you can switch off mid-workflow
Add white space to an image, one side at a time
Qwen editing, but with a mask so it leaves the rest alone
Qwen editing with six reference images and twelve parallel tries
Save images, but only when you mean it
Segment anything by describing it — in the cloud, no GPU needed
One string in, up to six strings out
A contact sheet that exactly fills the frame
Glue three strings together, exactly as written
Eight strings in, one string out, zero surprises
Join a list of strings back into one
When a string of digits needs to be a real number
Turn one string into a list, without the manual labour
Sixteen art styles behind one dropdown
Pick a branch and skip all the wasted work
A ten-way switch that only runs the case you picked
The ten-way switch, again — they ship both, and that's fine
Lazy branching that only runs what you pick
The little lazy switch for when five options is plenty
Stamp words on your images without leaving ComfyUI
A no-frills text box that plays nice with dynamic prompts
ComfyUI Assistant Node
A custom ComfyUI node that integrates with the OpenAI Assistants API.
Features
- Call OpenAI Assistant from within ComfyUI
- Supports dynamic model selection
- Input fields for API key, assistant ID, seed, and prompt text
Installation
Clone or download this repo into your ComfyUI/custom_nodes/ directory:
git clone https://github.com/pvlprk/comfyui-pvl-api-nodes