Plan F: AI Image Design Workflow (Posters / Brand)

Midjourney / Jimeng / Flux

Write prompts in this order — subject → scene → style → composition → detail constraints — and generate 4-8 candidates per round. Midjourney for artistry, Jimeng for Chinese scenes, Flux for accurate text.

💡For consistent subjects, fix the description or upload a reference; multiple rolls of the same prompt usually yield a keeper.

Img2Img / inpainting

Use image-to-image to recompose, swap backgrounds and fix details; inpainting repairs broken hands, faces and edges. Far faster than regenerating — the key to quality.

💡Repair broken hands/text in a targeted region instead of restarting the whole image.

Ideogram / Canva

Set headlines, slogans and prices with Ideogram (accurate glyphs) or lay out in Canva. AI-embedded text often garbles; let a proper tool handle typography.

💡Proofread every word; leave safe margins so platforms don't crop it.

Topaz / Real-ESRGAN

AI images are usually 1024-2048px; for print or high-res use, upscale with a super-resolution model and export the required spec (RGB/CMYK, 300dpi, PNG/PSD).

💡Denoise before upscaling to avoid amplified noise; check platform size specs for e-commerce.