Generate versioned images from a connected Text or linked-Brand prompt source, typed image references, and model-specific framing and output settings.
Where: Flowgen → Add → Image & Vector → Gen Image.
Key ideas
- Prompt copy is a connected source: Use a Text source node for flow-specific image instructions or bind the exact compatible field on the linked project Brand. For assistant/MCP-authored work, do not insert prose into arbitrary node properties.
- References are typed and ordered: Image references connect to the reference input; accepted count and order depend on the selected model. Extra references may be rejected or ignored, so inspect the model’s actual limit.
- Model settings vary: Aspect ratio, resolution, output format, negative instructions, variants, prompt expansion, steps, guidance, seed, acceleration, and person controls are exposed only when supported by the selected model.
- Results are versioned: Generated results can be compared in history and opened in the image lightbox; use Add to canvas or Paste to Replace only when that action matches the intended downstream use.
Steps
- Add Gen Image and connect a Text source to the prompt input, or connect the intended compatible field from the project-linked Brand.
- Connect reference images to the image reference input in the order the selected model should consider them. Give sources clear names and roles.
- Choose the model, then review the controls it exposes: aspect ratio, resolution, number of versions, output format, and any model-specific guidance or negative prompt options.
- Check the accepted reference count and exact quote before a paid run. Remove or reorder references that exceed the model’s current supported limit.
- Run only after the source wiring and cost are reviewed. Compare versions and open a result in the lightbox for full-size inspection.
- Connect the chosen image output to the next node, or use the node’s available Add to canvas/Paste to Replace action when you explicitly want to place or replace a canvas asset.
Card, fields, and inspector
- The card displays prompt/source state, model, result, run status, quote, and version history as supported by the selected model.
- The prompt content should be traceable to a connected Text source or the project’s linked Brand field; avoid copy hidden in unrelated node properties.
- The shared inspector exposes additional model-supported controls. A setting absent for one model is not guaranteed to be available for another.
- The image lightbox supports full-size review and the common result actions configured for this node.
Reference handling
- Connect actual image assets through typed reference inputs rather than putting image URLs or descriptions into prompt text.
- Check source order and the model’s accepted reference count after changing models.
- A reference image influences generation; it is not necessarily the same as the source image being edited. Use the correct edit/input role when the workflow should transform an image.
Tips
- State the intended outcome and constraints in the connected source; use the linked Brand for reusable identity rules and references.
- Make a small, bounded first run before scaling the variant count or using a higher-cost model.
Limitations and important notes
- Model catalog, reference limits, controls, availability, and prices change; trust the current picker and quote.
- An image output does not automatically replace the source asset or connect itself to another node.