HiDream O1 Image: lock this still before you buy motion (1cr)
HiDream O1 Image is the stills row. Animate after approval. Do not generate the pack inside the video model.
HiDream O1 Image is the stills row. Animate after approval. Do not generate the pack inside the video model.
The catalog description is the differentiator: pixel-native text-to-image that reasons before it draws. Unified model producing high-resolution images up to 2K, with optional subject references for personalization. 1 credit. Text-to-image. No audio. No duration. Aspects 1:1, 2:3, 3:2, 4:3, 3:4, 16:9, 9:16. requires_image is false — the reference is optional, not the job.
Reasoning is not motion. It is a better still.
It reasons, then it draws
HiDream O1 Image spends the pass on whether the brief coheres, then it puts pixels down, up to 2K. That is why it is worth using as the lock step before any I2V row: you get a still you can inspect for identity, not a 6-second blur of "close enough."
Write a complete scene: subject, setting, light, mood, palette, composition. Optional subject references are for personalization — a person or product you need to recur — not a substitute for a prompt. If you attach a reference, you are steering identity, not asking the model to animate.
The sibling HiDream O1 Image Edit is the change-one-thing row, also 1 credit. Jacket wrong, pack right: edit. Idea wrong: reroll O1 Image. Motion wrong: you should not be in motion yet.
HiDream's roster is O1 full and O1 Dev, generate and edit. This page is the full-weight stills generate. Text to image is the launcher. Best AI image generator is the ranking.
2K stills, then stop
"Up to 2K" is the listed high-resolution promise. There is no video duration to pick because there is no video. People still try to "just run it as a clip to see the character move." That skips the only cheap review this model offers: a 1-credit, 2K frame in Slack.
Set aspect to the eventual cut. 9:16 here if the reel is 9:16. Many image-to-video models inherit crop from the still. A 1:1 O1 lock becomes a 1:1 clip, and you will crop a reasoned composition to death.
Do not generate the pack inside Kling, Hailuo, or LTX "to save a step." Those rows bill video rates and invent motion in the same pass as the label. O1 already reasoned about the label. Let someone say yes.
Motion is a later row
When the still is approved:
- Product turn or camera move: image-to-video, start frame = this file.
- Same person, new pose: another O1 still or an edit, not a 10-second lottery.
- Talking: a lipsync row that wants a plate, which you now have.
One credit is the whole argument. Spend it until the still is boringly correct. Then spend a video rate on that file only.
FAQ
What does "reasons before it draws" change in the prompt?
Write a brief that can be checked: who, where, light, what must be on the pack. Do not send a vibe word and expect the reasoning pass to invent a brand system. The model still draws one still, at 1 credit, up to 2K.
Can I attach a subject reference and skip the text?
No. Optional subject references personalize; they do not replace the scene description. requires_image is false. A reference without a prompt is not this product's default path.
Is HiDream O1 Image a 2K video model?
No. Content type is image. 2K is the still. Animate after approval on a video row.
When do I use O1 Edit instead of generating another still?
When the approved still is the identity and one attribute is wrong. Edit keeps the plate. A new generate is a new drawing, even at 1 credit.