FLUX.2 [max], GPT Image 1.5, and Z-Image-Turbo were available by the end of December 2025. FLUX.2 [max] and GPT Image 1.5 offered hosted image generation and editing; Z-Image-Turbo provided downloadable text-to-image weights and an official demo. BFL release log, OpenAI launch, Z-Image release log
What are the key facts about these December 2025 image models?
| Field | FLUX.2 [max] | GPT Image 1.5 | Z-Image-Turbo |
|---|---|---|---|
| Developer | Black Forest Labs. Overview | OpenAI. Model reference | Tongyi-MAI's Z-Image team. Repository |
| Released | December 16, 2025. Release log | December 16, 2025. Launch announcement | November 26, 2025. Release log |
| Type | Image generation and editing. Overview | Image generation and editing. Model reference | Distilled text-to-image generation. Model card |
| Size or parameters | Not published in the cited variant documentation. Overview | Not published in the cited model reference. Model reference | 6B model parameters. Model card |
| License and access | Hosted service; no open weights for this variant. Overview | Hosted OpenAI API; no open weights documented. Model reference | Apache-2.0 weights and an official demo. Model card |
| Where it runs | BFL-hosted playground and API. Release log | OpenAI-hosted image endpoints. API reference | Official online demo or compatible self-managed hardware. Repository |
Which tasks suit FLUX.2 Max, GPT Image 1.5, and Z-Image-Turbo?
FLUX.2 [max]'s release announcement emphasizes editing consistency, reference-based composition, and grounding search.
For a December 2025 shortlist, those capabilities make it relevant to evaluate on briefs that depend on existing images or real-world context. Launch capabilities
GPT Image 1.5's model reference emphasizes instruction following and lists generation and editing endpoints.
It provides another hosted route to test when your deliverable requires both an initial image and subsequent changes. OpenAI model documentation
Z-Image-Turbo's model card emphasizes distilled generation, photographic imagery, and English and Chinese text rendering. Downloadable weights make it the self-managed option in this particular comparison.
That is a deployment distinction, not a claimed quality ranking. Model card
Build your shortlist around work you can describe and inspect. A poster, a product scene, and a landscape exercise different requirements.
A model that produces an appealing landscape has not thereby passed your poster's spelling and layout requirements.
For a compact trial, use a product scene with a short label. Specify a blue bottle, a pale tabletop, side lighting, and a visible word on the label.
Separate the required features from optional decorative details before submitting the prompt.
What deployment and comparison limits should you check?
The shortlist is limited to releases available by December 2025. The instructions below use current API documentation; verify present access conditions separately from historical release dates. BFL release log, Current BFL endpoint, OpenAI launch, Z-Image release log
FLUX.2 [max] has no published local weights in the cited variant documentation.
Other FLUX.2 variants have different deployment arrangements, so a downloadable member of the family does not establish local availability for [max]. Variant comparison
Likewise, GPT Image 1.5's documented image API is hosted. Its model reference does not offer downloadable weights or a parameter count.
A local client script sends a request to the service; it does not run that model on your GPU. Model reference
BFL's current overview lists up to eight reference images through the FLUX.2 [max] API and ten through the playground. Use the limit for the interface you select when preparing a reference-editing comparison. Current interface limits
Avoid a single unsourced speed or price column. Compare billed work at matching output settings, and separate generation, input processing, failed attempts, and human finishing time in your own project notes.
State when a measurement is yours.
How do you try these image models through their documented interfaces?
For FLUX.2 [max], start from the official BFL playground or create an API key through the developer platform. The [max] endpoint accepts a text prompt for generation and reference-image inputs for editing. Endpoint specification
With BFL_API_KEY configured in your environment, this example submits an original test brief:
curl --fail-with-body https://api.bfl.ai/v1/flux-2-max \
-H "x-key: $BFL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A blue glass bottle labeled WATER on a pale table, side lighting.",
"width": 1024,
"height": 1024
}'
Keep the returned job information and poll the returned polling_url with the x-key authentication header. BFL's guide waits for Ready, retrieves result.sample, and handles Error or Failed as failures.
Download the result promptly because the output URL is temporary. Polling workflow
For GPT Image 1.5, select the documented gpt-image-1.5 identifier when using OpenAI's image-generation endpoint.
For Z-Image-Turbo, follow the official repository's demo link or its published Diffusers workflow. OpenAI identifier, Z-Image access
For implementation detail, continue with the sibling FLUX.2 Max guide and Z-Image-Turbo online guide. Keep this comparison focused on deciding what to evaluate.
Use the same scene description for the first pass. Record the model, input prompt, output dimensions, and requested settings.
If one service rewrites or enhances your prompt, record that behavior rather than claiming the submitted briefs necessarily reached identical processing stages.
Review the results without service names visible if practical. Check the bottle's color, object shape, label spelling, and light direction separately.
Use a written scorecard, and retain failed examples alongside successful ones so the trial is not just a gallery of winners.
Then run a second task that reflects your actual need. For an editing project, define a specific change and list what must remain intact.
Compare only models and endpoints that document that operation; do not score an unsupported task as a visual-quality failure.
How should you choose between hosted editing and local generation?
| Priority | Candidate to investigate | Evidence to collect yourself |
|---|---|---|
| Reference-led hosted editing | FLUX.2 [max]. BFL overview | Whether the requested change preserves the required details. |
| Hosted generation and editing integration | GPT Image 1.5. Endpoint support | Whether the output and revision workflow fit your application. |
| Operating downloadable weights | Z-Image-Turbo. Model card | Whether your complete environment produces acceptable results. |
The ComfyUI complete guide helps frame the self-managed workflow option. Treat this table as a task-based shortlist; it does not claim a benchmark winner or substitute a model name for testing.
What are common questions about the December 2025 comparison?
Was FLUX.2 Max available in December 2025?
Black Forest Labs released FLUX.2 [max] on December 16, 2025. Its launch entry names the BFL playground and API as access paths. Release log
Can I run FLUX.2 Max locally?
FLUX.2 [max] is documented as a hosted playground and API model, without downloadable weights. Local deployment options elsewhere in the FLUX.2 family do not provide a local [max] checkpoint. Variant overview
Which has open weights: FLUX.2 Max, GPT Image 1.5, or Z-Image-Turbo?
Z-Image-Turbo publishes downloadable weights under Apache-2.0. FLUX.2 [max] and GPT Image 1.5 document hosted access without open-weight downloads. Z-Image card, BFL overview, OpenAI reference
How should I compare these three image models for a project?
Give FLUX.2 [max], GPT Image 1.5, and Z-Image-Turbo the same representative text-to-image brief and evaluate spelling, objects, layout, and finishing effort. For an editing task, compare the documented editing routes of FLUX.2 [max] and GPT Image 1.5 separately from Z-Image-Turbo's text-to-image workflow. BFL capabilities, OpenAI endpoints, Turbo model card
Sources
- Black Forest Labs dated release notes
- FLUX.2 model-family and deployment overview
- OpenAI GPT Image 1.5 model and snapshot reference
- OpenAI GPT Image 1.5 launch announcement
- Z-Image release history and official access routes
- Z-Image-Turbo official model card
- FLUX.2 Max endpoint specification
- BFL generation and polling workflow
Top comments (0)