Nodes/comfyui-stillfront/SF WaveSpeed Flux Kontext Pro
ComfyUI Node

SF WaveSpeed Flux Kontext Pro

The iterative editor for style swaps

By razvanmatei-sf·Created 9 months ago·Updated 7 months ago· 0
SF WaveSpeed Flux Kontext Pro
  • client
  • output_image
promptA woman is brewing tea
image_url
guidance_scale3.5
enable_sync_modetrue

Of all the Flux Kontext nodes in this pack, Pro is the one that feels like a scalpel. It's the API-only middle tier of Kontext - the one Black Forest Labs never released as weights - and the node that serves it is strikingly minimal: an image, a prompt, a guidance scale, done. No step count, no seed, no image count. That restraint is the feature. If you're chaining edits - "make it a painting," then "now put her in a rainy street," then "now it's dawn" - the lack of knobs means nothing distracts you from the actual instruction.

How it works

Same Kontext architecture as Dev, one tier up: your input image is concatenated into the model's context rather than injected as an adapter, so the subject is preserved structurally while your sentence does the editing. The node posts your prompt and image_url to WaveSpeed's flux-kontext-pro endpoint and returns a single IMAGE. Pro was the launch-tier workhorse of the Kontext lineup - it's what the "Flux Image Editing is Crazy" era was built on - and being API-served means you get the full-fidelity weights without a 24GB+ GPU in the room.

The inputs that matter

Three required fields plus the wiring:

  • client - from SF WaveSpeed Client. Non-negotiable.
  • prompt - the edit instruction. The default is "A woman is brewing tea," which is a decent hint that this model rewards scene descriptions, not abstract vibes.
  • image_url - the input image as a public URL. Same story as every Kontext node: the image must be fetchable by WaveSpeed's servers.
  • guidance_scale - default 3.5, range 1.0–20.0. Go up when the model drifts from your instruction; down when it over-edits.
  • enable_sync_mode - default true; leave it on unless you want to poll manually.

Output: output_image (IMAGE). Wire it into a preview, a save node, or chain it straight into the next Kontext node for iterative edits.

Installing it

In the SF ComfyUI Nodes pack from Stillfront. ComfyUI Manager → "SF ComfyUI Nodes" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Stillfront/comfyui-sf-nodes.git
cd comfyui-sf-nodes
pip install -r requirements.txt

Plus a WaveSpeed API key via SF WaveSpeed Client.

Common issues

The URL problem is king here: a non-public or login-walled image_url fails immediately, and it's the #1 reason a fresh workflow errors. Also note there's no seed and no step dial on this node - if you need reproducible edits or fine control over iteration count, that's what the Dev variant (with its extra knobs) is for. And the censorship caveat that follows Kontext everywhere applies: the model refuses some content at the weights level, and per BFL's licensing even running an NSFW-flavored workflow through a Kontext API tier is on shaky ground. For clean commercial-style editing - product shots, character re-styling, background swaps - it's hard to beat.

CategoryStillfront/WaveSpeed

Inputs (5)

NameTypeDefaultDescription
clientWAVESPEED_AI_API_CLIENT
promptSTRINGA woman is brewing teaText prompt describing the desired transformation or scene
image_urlSTRINGThe image URL to transform (connect from Upload Image node)
guidance_scaleFLOAT3.51–20How closely to follow the prompt (1.0 = loose, 20.0 = strict)
enable_sync_modeBOOLEANtrueWait for image generation to complete before returning

Outputs (1)

NameTypeDescription
output_imageIMAGE