FLUX Kontext Max is the premium tier of Black Forest Labs' FLUX.1 Kontext family — a flow-matching (rectified-flow transformer) model that unifies instruction-based image editing with text-to-image generation. It shares the same in-context editing interface as FLUX Kontext Pro but allocates more compute to prompt adherence and typography, so it lands harder on tight instructions and on text rendered inside an image.
Hand it a reference image plus an instruction — swap an object, change a background, replace a label — and it edits only what you name while preserving faces, layout, and lighting across multiple turns. It is not a language model: there is no token context window, no document analysis, and no RAG. Images and text go in; an image comes out.
| Field | FLUX Kontext Max |
|---|---|
| Maker | Black Forest Labs (BFL) |
| Family / tier | FLUX.1 Kontext — [max] (premium) |
| Architecture | Flow matching / rectified-flow transformer |
| Positioning vs Pro | Higher prompt adherence + best in-family typography |
| Tasks | Image editing (local & global) · text-to-image · character reference · style reference · in-image text editing |
| Inputs | Reference image + text instruction (text-only for generation) |
| Output | Image, ~1 MP (1024×1024 class) |
| Prompt length | up to ~512 tokens (per BFL partner docs) |
| Multi-turn | Yes — iterative refinement |
| Open weights | No — [max] is API-only; only [dev] is open-weight (non-commercial) |
| Released | May 29, 2025 |
| GPTProto model string | flux-kontext-max |
| GPTProto endpoint | https://gptproto.com/v1/images/generations |
| GPTProto price | $0.064 / image (official BFL list $0.08) |
On GPTProto this page defaults to the image-edit task — a reference image plus an instruction in, an edited image out, at $0.064 per image. Text-to-image (prompt only, no reference image) is also supported via the task selector.
FLUX Kontext Max vs Pro vs gpt-image-2
Same interface, same tasks, same character consistency. The split is fidelity versus speed-and-cost — and both undercut the official list price on GPTProto.
| FLUX Kontext Max | FLUX Kontext Pro | |
|---|---|---|
| GPTProto price | $0.064 / image | $0.032 / image |
| Official (BFL) list | $0.08 / image | $0.04 / image |
| Prompt adherence | strongest | strong |
| In-image typography | best in family | strong |
| Speed | fast (a touch behind Pro) | fastest tier |
| Editing + character consistency | yes | yes |
hoose Max when in-image text or exact prompt adherence is the deciding factor — marketing copy on a poster, a relabeled product shot, a headline swap. Choose Pro when speed and cost lead and the edits are straightforward. If you're comparing against the OpenAI stack, gpt-image-2 is on GPTProto too; on editing tasks FLUX Kontext tends to hold character identity and in-image text better and returns in under ten seconds, where GPT-Image editing often runs longer.
Switching from Black Forest Labs (drop-in)
Moving from BFL's direct API to the FLUX Kontext Max API on GPTProto is a small diff, not a rewrite:
1. base_url → https://gptproto.com/v1/images/generations
2. auth → Authorization: Bearer YOUR_GPTPROTO_KEY (replaces BFL's x-key)
3. model → keep "flux-kontext-max"
The same GPTProto balance then covers FLUX Kontext Pro, Seedream, gpt-image-2 and 200+ other models — one key, one invoice, no separate BFL credit top-ups and no per-provider sign-up.
Prompt tips for FLUX Kontext Max (typography-first)
Max earns its price on text and precise instructions, so prompt for that:
• Text replacement: use quotes — replace 'OLD' with 'NEW'.
• Preserve styling: add "keep the same font style, color, and position".
• Name the subject, not a pronoun: "the man in the red jacket", not "him".
• State what must NOT change: "…while keeping the background unchanged".
• Be specific about style: "convert to watercolor", not "make it artistic".
One edit per turn works best; prompts cap around 512 tokens, and quality can drift after several consecutive edits — restart from a clean base image if artifacts appear.
