Images & Interactive Answers
Generate or edit pictures in a conversation, and explore explanations with interactive controls.
Set up image generation
- Open Settings → Models & Providers → Image generation.
- Choose a ChatGPT plan account, an OpenAI API account, or an image-capable compatible account. Use Add Account… if none is available.
- For an API or compatible account, choose a model, default size, and quality. For Other…, enter the model name and commit it by pressing Return or leaving the field.
- Check the status row to confirm the local agent accepted the settings.
None (off) disables image generation. The image account is selected separately from the conversation account. Subscription and API routes do not fall back to one another.
Image models in 3.0
| Image route | Available controls |
|---|---|
| ChatGPT plan | GPT Image 2 through the managed helper. Size and quality remain automatic; requested preferences can be described in the prompt. Masks are unavailable. |
| OpenAI API / compatible account | Model picker including GPT Image 2.5 Sunburst, GPT Image 2.5 Flare, GPT Image 2, GPT Image 1.5, GPT Image 1, and GPT Image 1 mini, plus custom model names. Provider support is required. |
New API image settings default to Sunburst. Existing saved choices survive an update. The 2.5 models add Extra High and Max quality. GPT Image 2 and 2.5 expose 2K/4K presets and validated custom dimensions; unsupported choices reset to Automatic when switching models. Invalid custom values preserve the last valid saved size.
Custom dimensions use multiples of 16, at most 3,840 pixels per edge, an aspect ratio between 1:3 and 3:1, and 655,360–8,294,400 total pixels. Higher-resolution options above 2,560 × 1,440 remain experimental. The model picker is not a guarantee of account eligibility. Live image-generation and recovery campaigns remain deferred.
Create a picture
Open a Work conversation. Ask for the image you want and describe the subject, composition, style, and intended use. For example: “Create a square illustration of a quiet mountain cabin for the cover of this project.”
Outside Bypass mode, generation asks you to approve the prompt, model, size, quality, account, provider host, and destination unless you already allowed generation for that session. The prompt is sent to the selected image provider. Generation is limited to four images per turn and 24 per session in every permission mode. Unattended schedules, event triggers, and workflows cannot generate images. A successful result appears as an image card, is saved under Locus Images in the workspace, and is captured in Library → Outputs.
The image card offers Copy Image, Save As…, and Edit in chat. These actions are also available for supported inline workspace images. Use Stop to interrupt a generation while the provider is still working.
Edit an existing image
Choose Edit in chat on an image. Locus attaches the source and prefills a request with its workspace path. Add the change you want, then review and send the draft. You can also attach an image and describe an edit in a new request.
Image edits always require approval. The source image and prompt are sent to the selected image provider. The original reference remains available; review the generated result before using or sharing it.
Changing the image account or its settings updates the main agent and chat workers. A change refused during an active turn is retried after the turn ends. The API key is handed to the local agent in memory, rather than written into its configuration.
Use an interactive explanation
An answer can include an interactive chart, diagram, or small explanatory tool. The widget runs in a sealed offline view and includes a written summary. It does not require an image account.
Use Open larger, Copy HTML, or Save As… on the card. Render interactive answers, under the same settings section, controls whether these widgets render. Mobile, exports, and conversations with rendering switched off use the written summary.
Save and share
Markdown, plain-text, and PDF chat exports can include generated images and the interactive HTML beside the exported conversation. Review the exported files before sharing. Library, Documents & Outputs keeps captured versions and offers preview, export, comparison, and revision actions.
Image tools require an interactive Work task with network and workspace-write access. Plan, conversation-only Chat/Ask, private Identity tasks, specialists, helpers, and read-only agents cannot use them. Changing account settings does not raise those permissions. A ChatGPT image task uses an isolated image-only helper session, and the ordinary conversation cannot bypass Locus's image permission controls.