ChatCut can generate a new still from text or revise an image using references, then add the result to My Assets for timeline editing. As reviewed on 28 October 2026, the inspected integration exposes three GPT Image selectors and Nano Banana Pro. Model documentation, tool controls and completed endpoint tests are separate evidence; the dated table below does not certify deployed production.
Which image model and route does this describe?
The table covers ChatCut-hosted generation. An external agent’s own image tool uses that host’s access and allowance. Check the active tool and confirmation before submitting; a conversation in Codex can also call ChatCut generation and consume ChatCut credits.
| Model and public ID | Capabilities, integration and evidence |
|---|---|
GPT Image 2gpt-image-2Availability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Text generation and image edits; direct API documents up to 16 references and a 32,000-character prompt ceiling. ChatCut integration: Exact selector exposed; 10 combined references; no background control for Image 2. Endpoint evidence: One low-quality 1024×640 PNG succeeded. Transparent background explicitly rejected; high-resolution boundary outputs unverified. |
GPT Image 2.5 Flaregpt-image-2.5-flareAvailability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Generation and edits; documented quality and transparent-output controls; custom sizes follow pixel/edge rules. ChatCut integration: Exact selector, 10 combined references and Image 2.5 background control exposed. Endpoint evidence: One low-quality 1024×640 PNG succeeded. Transparent output and high-resolution boundaries unverified in endpoint audit. |
GPT Image 2.5 Sunburstgpt-image-2.5-sunburstAvailability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Generation and edits; documented quality and transparent-output controls; custom sizes follow pixel/edge rules. ChatCut integration: Exact selector, 10 combined references and Image 2.5 background control exposed. Endpoint evidence: One low-quality 1024×640 PNG succeeded. Transparent output and high-resolution boundaries unverified in endpoint audit; separate local scene run produced larger stills. |
Nano Banana Progemini-3-pro-imageAvailability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Text/image generation and edits; 1K/2K/4K tiers; up to 14 image references documented. ChatCut integration: nano-banana selects Pro; local reference cap 14 and 1K/2K/4K controls exposed. Endpoint evidence: Final 1024×1024 PNG succeeded at 1K. 2K/4K and full-reference boundaries unverified. Saved adapter replay selects an intermediate thought image. |
Nano Banana 2gemini-3.1-flash-imageAvailability: Pinned integration and saved dev checks; production unverified. Source links describe native specifications. Reviewed: | Native documentation: Distinct model; documented 512/1K/2K/4K tiers and up to 14 images. ChatCut integration: No separate selector in pinned ChatCut integration; nano-banana continues to select Pro. Endpoint evidence: Final 512×512 PNG succeeded in a direct development check. This does not establish a ChatCut selector or full-boundary generation. |
GPT Image 2.5 Flare, Sunburst and Image 2
The exact aliases are gpt-image-2.5-flare, gpt-image-2.5-sunburst and gpt-image-2. OpenAI positions Flare around speed and Sunburst around precise edits; there is no controlled comparison in this review. Use an available model and one small representative draft rather than assuming a quality or timing advantage. All three have a 10 combined-reference ChatCut cap; the direct API’s documented 16-image limit does not raise that cap.
Transparency: the checked endpoint explicitly rejected transparent backgrounds for Image 2. The inspected integration exposes the background option only for Image 2.5, but the audit did not complete a transparent 2.5 result. Inspect the output’s alpha before relying on it. For prompt limits, output pixel rules and edit controls, read the GPT Image reference.
Nano Banana Pro versus Nano Banana 2
In the inspected integration, nano-banana maps to Nano Banana Pro / gemini-3-pro-image, with 1K/2K/4K controls and a local 14-reference cap. Nano Banana 2 is gemini-3.1-flash-image; a small direct development output succeeded, but there is no separate selector in the pinned integration. Do not use the Pro alias to request Nano Banana 2.
Pro’s 2K/4K tiers are documented; malformed-input probes did not prove higher-tier or full-reference generation. The saved Pro response also exposes an adapter bug: an intermediate thought image can be selected before the final image. Inspect the delivered asset even when the request succeeds. For a product sheet, begin with one clear product reference and review silhouette, text and composition manually. No matched Pro-versus-2 quality or cost comparison is supplied. See the Nano Banana reference.
Prepare the prompt and references
- Decide the subject, composition, ratio, requested size, image count and where the still belongs. Text-only generation needs no uploaded reference.
- To preserve a person, product or layout, attach clear references with +, or add project context with Selection Mode and
@. Describe each reference’s role and what must stay fixed; references do not lock identity or pixels. - For GPT, prepare PNG/JPEG/WebP and keep the prompt at 32,000 characters or fewer. Native file limits differ from attachment transport: ChatCut’s public-URL references have a 5 MiB limit. Rasterize SVG before submitting; see generation references.
- Confirm your current plan, balance and available model. The inspected ChatCut image workflow requires an active Pro plan and credits. A host agent’s own tool has a separate allowance.
Using the ChatCut GPT Image 2.5 Sunburst option if available, generate one 16:9 product still at 1K with an opaque background. Use @product-photo as the product reference. Keep its silhouette and colors, place it on a light-gray surface with soft light from the left, reserve empty space on the right, and add no text.
Generate, inspect and place the still
Choose the actual model
Name the ChatCut model shown in your build and ask for one draft. Specify the ratio, requested tier, quality and reference roles. Request a transparent background only with an exposed Image 2.5 control, with the completion caveat above.
Review and submit
If a generation confirmation appears, check the model, settings and credit estimate, then select Allow. Follow progress in the conversation or My Assets. A successful job still needs visual inspection.
Inspect the output
Check actual dimensions, product details, text, composition and any alpha channel. For Nano Banana Pro, confirm that the asset is a final image. Reject an unsuitable draft before generating a larger batch.
Place and revise
Drag the completed asset from My Assets to a video track. Adjust crop, position, scale and duration. To revise it, reference the latest accepted image and state both the change and the details to preserve.
The inspected ChatCut workflow defaults to one image and accepts 1–10 requested outputs. It submits separate jobs, assets and charges for each image. These are output counts, separate from the model’s reference-image cap. Results are raster stills; generated objects do not become separate editable layers. If an image already exists, use My Assets → Import → Files… and place it directly.
Which settings produced the local casting and plate examples?
In a 27 October local scene test, reviewed on 28 October, Sunburst used high quality and requested 16:9: 20 casting sheets at requested 2K returned 2304×1296, and 10 office plates at requested 4K returned 3776×2128. The settled still-image charge was 4.00 credits per group, 8.00 total. This demonstrates those local Sunburst outputs; it is not a Flare comparison or a production price. Production cohort/catalog equivalence and new quotes remain unverified; this example has no video input or output duration and supplies no reusable rounding rule. The wider scene run’s 22.24 Agent credits were separate from media.
For your request, check the current estimate at matching model, tier, quality, ratio, count and reference settings, then open Credits history → Detail for the settled charge. Credits Policy owns current rates; generation credits explains estimates. The older Nano Banana card figure names Nano Banana 2 and cannot be treated as a Pro price.
Frequently asked questions
Which ChatCut aliases are compared for GPT Image?
What do Nano Banana Pro and Nano Banana 2 map to?
What has actually been tested at 2K and 4K?
Which transparency controls work on the checked route?
Does generating images inside Codex always use ChatCut credits?
Last verified: 28 October 2026. Official sources, accessed 27 October: OpenAI Image 2, Flare, Sunburst, edit API, Google image guide, Pro model card and Nano Banana 2 card. Pinned integration, saved checks and the local scene ledger were reviewed on 28 October; no new generation or production verification was performed. For missing options or failures, use workflow failures, generation confirmations and the model matrix.