Image generation: models, modes and prompt tips

Open Image in the sidebar and choose a model at the top:

• Flux: realistic lighting and detail, the all-round default.
• Qwen: strong Chinese prompt understanding, good for posters and text layouts, supports multi-image references.
• PonyXL 3D: anime and 3D-render styles.

Supported modes and ratios differ per model; the panel refreshes when you switch.

• Text-to-image: prompt only.
• Image-to-image: one reference image + prompt; keeps the composition, changes style or content.
• Multi-image: several references at once (limit shown on the slots), e.g. put person A into scene B.
• Face swap: one face photo + one target image, no prompt. The face must be a clear frontal photo of a real person; animals, profiles and cartoons are rejected upstream.
• Motion control: one character image + one pose reference to restage the character.

A reliable structure: subject + action / expression + setting + lighting + style + quality.

Example: an orange cat lying on a beach in the afternoon, squinting in the sun, glittering sea, soft backlight, film look, high detail.

• Not sure what to write? Click Expand on a text node or in the studio and AI turns one sentence into a full description.
• Put what you don't want in the negative prompt: blurry, extra fingers, text, watermark.
• Chinese and English prompts both work.

• Ratio sets the output size: 3:4 / 9:16 for posters, 1:1 for avatars, 16:9 for banners.
• The maximum count per run depends on the model; credits are charged per image.
• The cost is shown before you submit; with insufficient balance nothing is charged.
• Failed generations are refunded automatically.

Generated images go straight to the Library: download, favourite, set as cover, or drag them onto a canvas for image-to-image, face swap or video.

Library storage depends on your plan; free up space by deleting results you no longer need.