Skip to main content
Generation time depends on server load. Image generation typically completes in a few seconds. Video generation takes longer — Enhanced Video longest of all. If a generation is still running after five minutes, refresh the page.
Tokens are returned automatically for failed generations. Wait a few minutes and check your balance. If tokens haven’t been restored after ten minutes, contact support with the approximate time of the failed generation.
Three common causes: Prompt Enhancer is rewriting your prompt, the model is misinterpreting token order, or conflicting terms are cancelling each other out. Try toggling Prompt Enhancer off in the settings panel, put your most important terms first, and remove vague quality descriptors like “beautiful” or “perfect”.
Check the resolution setting — a low resolution produces soft output regardless of prompt quality. If resolution is correct, add quality terms to the prompt: sharp focus, high detail, 8k. For video, run the clip through Enhanced Video. See Enhanced Video.
This is a known limitation of diffusion models. Add deformed hands, extra fingers, missing fingers, bad anatomy to your negative prompt. For anatomy-specific improvements, use the Better Breast or BetterCock LoRA. See LoRAs.
The motion prompt doesn’t contain explicit movement terms. Add camera and body motion terms: slow pan right, walking forward, hair blowing, slow zoom in. See Video Prompting.
Image-to-Video only supports images generated with the Realism model. If your image was generated with another model, regenerate the first frame using Realism. Also check that the resolution matches between the source image and the video settings. See Image-to-Video.
Some content falls outside what the platform will generate. See Content Restrictions for details.