comfyui-stillfront

One node, every frontier LLM — Claude, Gemini, and GPT-5 right in your graph

Up to 50 text boxes, one comma-joined string

The Qwen aspect-ratio cheat sheet — pick a preset, get real pixels and a preview

Word count, a token estimate, and a summary string — the LLM cost sanity check

Imagen 3 in ComfyUI — Google's text-to-image, minus the ImageFX tab

Google's Imagen 4 straight into ComfyUI, billed to your GCP project

Imagen 4 upscaling without a GPU — but know the 17-megapixel ceiling

Nano Banana Pro on Vertex AI — the closed Gemini image model, minus the app

Nano Banana Pro editing in ComfyUI — up to 14 reference images, thinking mode included

Animate a still into a Veo 3.1 clip, right inside a ComfyUI graph

Veo 3.1 video with native audio — the one thing open source still can't do

The key holder for every WaveSpeed node — wire this in first or nothing runs

Flux with real spatial control, served over the WaveSpeed API

The open-weight Kontext tier, without the 24GB of VRAM

Kontext Max, the top editing tier — better typography, fewer dials

The iterative editor for style swaps

A 4K-or-8K generative upscaler that runs entirely in someone else's cloud

Talking-head video from one portrait and an audio file

Two characters, two audio clips, one image — an actual animated conversation

Edit an image by describing the change — the closed Gemini editor, by API

Nano Banana Pro Edit — the community favorite edit model, in your graph

Two Nano Banana Pro edits per run from the same prompt

Nano Banana Pro Edit at 4K or 8K — the big-canvas tier

Nano Banana Pro text-to-image, up to 4K, billed per call

It always makes exactly two

The 8K tier of Nano Banana Pro text-to-image

The boring, reliable way to get Nano Banana into ComfyUI — one prompt, one image

Qwen-Image-Edit — bilingual image editing that keeps the style while it changes the content

Qwen-Image-Edit with LoRAs — host the edit, bring your own style

Qwen Edit with up to three reference images — consistency across a whole batch

Three reference images, three LoRA slots, zero VRAM

Qwen-Image text-to-image from a prompt — the pack's bilingual workhorse

Qwen-Image generation with up to two hosted LoRAs — style without the download

Video upscaling with no GPU, no ffmpeg, no patience

Seedream V4 — ByteDance's flagship image model, rented by the prompt

Seedream V4 editing through an API, with a 4096px preset your local card can only dream about

Seedream V4 Edit Sequential in one run

Seedream V4, but for storyboards — one prompt, up to 15 sequential images

Sora 2 image-to-video — animate a still you already like, without a GPU

The price tag is the feature

Sora 2 text-to-video from a prompt — scene, camera, and audio cues in one box

Sora 2 in Pro resolution — the upgrade when 720p isn't going to cut it

VEO 3.1 image-to-video with zero local GPU — the fast lane

Veo 3.1 Fast without a Google Cloud project — and yes, it does native audio

A target frame

The same character, every frame

Pure text-to-video on VEO 3.1, 1080p, with native audio — no GPU required

Put a character into someone else's video, no GPU farm required

Wan 2.2 I2V with a start AND end frame — the cloud node that anchors both ends

Wan 2.5 image editing, done by instruction — and it returns a real IMAGE

Animate a still with Wan 2.5 — and optionally drive the video with an audio clip

Cheaper, quicker, good enough

Wan 2.5 text-to-image over the API, straight into your graph as a tensor

API-only, and this node is the API

Type it, render it, move on