Reference Image Mode
An AI image generation control mode that encodes a reference image as a separate identity or style vector and injects it in parallel with the text prompt without binding the composition, thereby preserving facial and clothing features in entirely new scenes.
Timeline (last 90 days)
When making turnaround views, you should input the main reference image separately in reference-image mode for each angle, generate each angle one by one, and manually select consistent images to form the reference image library.
Reference image modes (such as Midjourney's --cref, Stable Diffusion's IP-Adapter and Flux's Redux) encode the reference image into an independent identity/style vector that is injected in parallel with the text prompt and is not bound to composition.
When making turnaround views, reference image mode should be preferred over high-denoise image-to-image.
All Facts (3)
When making turnaround views, you should input the main reference image separately in reference-image mode for each angle, generate each angle one by one, and manually select consistent images to form the reference image library.
50%UnverifiedReference image modes (such as Midjourney's --cref, Stable Diffusion's IP-Adapter and Flux's Redux) encode the reference image into an independent identity/style vector that is injected in parallel with the text prompt and is not bound to composition.
50%UnverifiedWhen making turnaround views, reference image mode should be preferred over high-denoise image-to-image.
50%