SF WaveSpeed Flux Kontext Max

If Pro is the scalpel and Dev is the workbench, Kontext Max is the boutique tier - Black Forest Labs labeled it "experimental" when it launched, and it's the one with the strongest prompt adherence and the best typography. SF WaveSpeed Flux Kontext Max serves that tier over the WaveSpeed API, and like the Pro node it keeps the interface brutally simple: image, prompt, guidance, and one extra knob Pro doesn't have - a safety_tolerance dial.

How it works

Max uses the same in-context editing architecture as the rest of the Kontext family - your input image gets concatenated into the model's context so the identity is held structurally while the prompt does the work. What's different is the serving tier: Max is the API-only flagship, the one BFL claimed would combine maximum editing consistency with better text rendering. The node posts your prompt and image_url to WaveSpeed's flux-kontext-max endpoint and returns a single IMAGE. The default prompt, "To toy style," hints at what people actually use it for - dramatic stylization jobs where Pro's consistency and Max's prompt adherence both matter.

The inputs that matter

• client - from SF WaveSpeed Client. As always, wire it first.
• prompt - the transformation or scene instruction.
• image_url - the input image as a public URL. The usual requirement: it must be fetchable by WaveSpeed's servers.
• guidance_scale - default 3.5, 1.0–20.0. Higher = stricter adherence to the prompt.
• safety_tolerance - a 1–5 scale (default 2, with 1 strictest). This is the knob the other Kontext nodes don't have. It tunes how permissive the model's content filtering is - handy when a legitimate prompt is getting bounced, and a sign that the model still carries BFL's safety baggage regardless of what you set.
• enable_sync_mode - default true.

Output: output_image (IMAGE).

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 key via SF WaveSpeed Client.

Common issues

Max inherits the Kontext family's two cautions: the input image_url has to be publicly reachable or the call fails at the first hop, and the model's censorship is baked in at the weights level - the safety_tolerance dial widens or narrows the window, it doesn't remove the wall. One more thing specific to Max: it's the highest-tier Kontext option in this pack, which usually means the priciest per call. Use it for the jobs where typography and prompt adherence are actually the bottleneck; for rapid iteration, Dev at 28 steps will get you 90% of the way for a fraction of the spend.