Generate and edit images with references, preserving characters, outfits, and IP across scenes, with strong support for text-heavy layouts and posters.
Text-to-Image creates from a prompt. Reference/Edit uses uploaded images to edit or preserve a product, character, outfit, or IP across scenes.
Describe the image to generate or the edit to apply. HiDream-O1-Image is strong at references, subject consistency, poster layouts, and multilingual copy.
Sign in to spend tokens
Prompt / Edit Instruction is required
Your generated images will appear here
HiDream-O1-Image generates and edits images from references, keeping the same product, character, outfit, or IP consistent across scenes — with strong support for text-heavy layouts and posters. Prizmad ships it as a built-in tool: pick a mode, upload references, generate.
HiDream-O1-Image is an image model built around references and consistency. Instead of generating a new random subject every time, it can take up to four uploaded reference images — a product, a character, an outfit, a brand asset — and preserve that subject across new scenes, poses, and compositions. That makes it particularly useful for ad work, where the same product or presenter needs to appear in many creatives without drifting off-model.
It runs in two modes. Text-to-Image generates from a prompt alone, with strengths in poster layouts, subject consistency, and multilingual copy. Reference/Edit uses your uploaded images to edit a scene or drive subject-consistent generation — keep the same model face and outfit in a new campaign scene, or place your exact packaging into a luxury poster composition. When you upload a single reference, a Keep Original Aspect Ratio switch asks the model to preserve its proportions; otherwise you pick from 9:16, 16:9, 1:1, and 4:5, and can generate 1–4 images per run.
On Prizmad, HiDream-O1-Image uses the same token balance as AI avatars, video, voiceover, captions, and music. Build a consistent creative series around one product photo, then push the winning image into a video tool on the same account — one balance, one library, full commercial rights.
Place the exact same product — packaging, label, and logo preserved — into multiple ad scenes so your whole campaign stays on-model.
Poster-style creatives with clean editorial layouts and readable headline text for product launches and promos.
Keep the same model face and outfit from a reference photo across new fashion or lifestyle campaign scenes.
Multilingual campaign copy inside the layout, so one visual concept can ship across several markets.
Cinematic hero visuals with bold typography and high-end product lighting for store pages and marketplace listings.
Preserves characters, outfits, products, and IP from your reference images while changing the scene, pose, or composition around them.
Reference/Edit mode takes up to 4 uploaded images and applies your prompt as an edit — new background, new scene, new layout — while keeping the subject intact.
Strong at poster-style compositions with readable headline text — launch posters, campaign layouts, and editorial-style product pages.
Handles multilingual text in layouts, which is useful when the same creative needs to ship in several markets.
Generate 1–4 images per run to compare variations of the same concept and pick the strongest creative.
Outputs in 9:16, 16:9, 1:1, or 4:5 — or preserves the original aspect ratio of a single uploaded reference when the switch is enabled.
Pick a mode: Text-to-Image to generate from a prompt alone, or Reference/Edit to work from uploaded images.
In Reference/Edit mode, upload up to 4 reference images — your product, character, outfit, or brand/IP assets. The first thing you want preserved should be clearly visible in the references.
Write your prompt or edit instruction. Describe the scene, layout, lighting, and any headline text; tell the model explicitly what to keep from the references (for example, "preserve the packaging and logo").
Choose an aspect ratio — 9:16 for Stories and Reels, 1:1 for feed, 16:9 for banners, 4:5 for portrait feed posts. With a single reference you can instead enable Keep Original Aspect Ratio.
Set the number of images (1–4) and click Generate. Results typically arrive in about 1–5 minutes.
Review the variations in your asset library, download the winner, or iterate with a refined prompt while reusing the same references.
Each HiDream-O1-Image run costs 2 tokens, in both Text-to-Image and Reference/Edit modes. Tokens come from your token balance.
HiDream-O1-Image is unlocked on the same token balance you use for AI avatars, video, voiceover, and music — no separate model account, no API key setup, no extra billing.
If your tokens run dry mid-campaign, buy a one-off top-up directly from the top-up page. Generated images stay yours with full commercial rights.
It's available to any account with enough tokens rather than offered as a standalone free tool. Each generation costs 2 tokens from your token balance, and one-off token top-ups are available if you run out.
A flat 2 tokens per generation in either mode. There's no per-image licensing fee and no separate billing — it draws from the same token balance as every other Prizmad tool.
Yes. Images generated with HiDream-O1-Image on Prizmad are yours to use commercially — in paid ads, on landing pages, in listings, and in print. Just make sure you hold the rights to any reference images you upload.
Up to 4 in Reference/Edit mode. References can carry a product, a character, an outfit, or brand/IP details. When you upload exactly one reference, you can also enable Keep Original Aspect Ratio to preserve its proportions in the output.
Its edge is consistency from references — keeping the same product, character, or outfit recognizable across many scenes — plus text-heavy poster layouts and multilingual copy. If you need exact rendered text as the top priority, ChatGPT Image 2 is also available on the same token balance; if you need editable vector output, use Recraft V4.1.
Yes. Reference/Edit mode applies your prompt as an edit instruction over the uploaded references — for example, editing your product photo into a luxury poster scene while preserving the packaging and logo.
You choose 1 to 4 images per generation, and results typically arrive in about 1–5 minutes. Jobs run asynchronously, so you can queue a generation and keep working while it completes.
9:16, 16:9, 1:1, and 4:5 — covering Stories/Reels, banners, square feed posts, and portrait feed posts. With a single reference image you can alternatively keep its original aspect ratio.
Published 2026-07-16 · Last updated 2026-07-16