Choose the fast Flare or precise Sunburst GPT Image 2.5 route, set one of the supported canvas shapes and 1K–4K output tiers, connect up to sixteen references for an edit, and understand exactly when the edit model and per-reference Sparks are selected.
Where: Flowgen → Add → Image → Image Generation or Image Editing → Model.
Key ideas
- Two intent tiers: GPT Image 2.5 Flare is the fast route for iteration; Sunburst is the precise route for work where instruction adherence and refinement matter more than turnaround.
- Create and edit share one picker choice: The visible base model remains Flare or Sunburst. Connecting one or more image references automatically routes the run to its edit twin instead of requiring a second model choice.
- Resolution and references affect the quote: Base pricing is 8 Sparks at 1K, 9 at 2K, and 14 at 4K. Edit runs include the first reference; each additional reference adds 1 Spark, up to sixteen references total.
Steps
- Add an Image Generation or Image Editing node and open its card or properties.
- Choose GPT Image 2.5 Flare (fast) for rapid exploration or GPT Image 2.5 Sunburst (precise) for a more exact final pass.
- Write the requested subject, composition, visual treatment, and constraints in the prompt.
- Choose 1K, 2K, or 4K after balancing output detail against the visible 8, 9, or 14 Spark base quote.
- Choose Auto or a supported ratio: 1:1, 3:2, 2:3, 16:9, 9:16, 4:3, 3:4, 21:9, 27:16, 16:27, 9:8, or 8:9.
- For an edit, connect or attach up to sixteen image references and say what each one contributes—identity, product, pose, palette, composition, or style.
- Review the final quote: the first reference is included and references two through sixteen add 1 Spark each.
- Run the node, inspect the generated version, and preserve the same reference roles when refining so an identity or product does not drift between passes.
Model ids behind the picker
- gpt-image-2.5-flare and gpt-image-2.5-sunburst are the visible create routes.
- gpt-image-2.5-flare-edit and gpt-image-2.5-sunburst-edit are selected automatically when references are connected.
- All four ids use the same 1K, 2K, and 4K base price table.
Reference accounting
- Zero references is a create run.
- One or more references makes the request an edit run.
- The first reference is included in the base price; every reference after it adds 1 Spark.
- The hard reference cap is sixteen, so the maximum reference surcharge is 15 Sparks.
Tips
- Use Flare to explore prompt wording and composition, then switch the same node to Sunburst for the precision pass.
- Give every reference a distinct job in the prompt; a large unlabeled pile makes conflicts harder for the model to resolve.
Limitations and important notes
- The supported aspect-ratio list is model-specific; arbitrary ratios are not accepted by this route.
- A higher resolution increases pixel dimensions, not guaranteed semantic accuracy.
- References beyond sixteen must be removed or consolidated before the run.
Troubleshooting
The quote rose after adding references.
The first reference is included, while each additional reference costs 1 Spark. Remove references that do not have a distinct role.
The result follows the wrong reference.
State which image supplies identity, product, layout, or style, and remove contradictory references before rerunning.