Skip to content
DreaminaHub

Generating images in Dreamina

Three models at three price points. The difference between an efficient session and an expensive one is entirely about which model you use at which stage.

Last reviewed

Short answer

How does Dreamina image generation work?

Write a prompt, pick a model, set the aspect ratio, generate. The model picker is the only control that changes the price, from 2 credits on Image 3.0 to 6 on Seedream 4.0. Reference images anchor composition and character; inpainting fixes a region without paying for a whole new render.

Models
3
Cheapest
2 credits
Best quality
Seedream 4.0
Free per day
112 drafts

Models

What each one is for

Not a quality ladder to climb — a set of tools for different stages of the same job.

ModelCreditsPer day freeStrengthUse at this stage
Image 3.02112Everyday text-to-image. Fastest turnaround, lowest cost per render.Drafts, thumbnails, bulk concepting
Image 3.0 Pro456Sharper detail and noticeably better text rendering inside images.Posters, product shots, anything with words in it
Seedream 4.0637Highest fidelity, strongest prompt adherence, best at photoreal skin.Portraits, final deliverables, client work
Dreamina model selector dropdown showing the available image models
The model picker. This dropdown is the difference between a 2-credit render and a 6-credit one — check it before every generation, because it persists from your last session.

Walkthrough

From prompt to finished image

  1. 1

    Set the aspect ratio first

    Free
    Ratio cannot be changed after generation — re-cropping means paying for a second render. Decide the destination before you write the prompt.
  2. 2

    Draft on the cheapest model

    2 credits
    You are testing whether the prompt produces the composition you imagined. Fidelity is irrelevant at this stage and paying for it is wasted.
  3. 3

    Change one clause and re-run

    2 credits each
    Single-variable iteration. Rewriting the whole prompt loses the parts that were working and tells you nothing about which change helped.
  4. 4

    Add a reference image if consistency matters

    Free to attach
    For a recurring character or a specific product, a reference anchors identity in a way prompt text cannot.
  5. 5

    Final render on the best model

    6 credits
    Same prompt, higher-fidelity model, once. This is the only render that needs to be expensive.
  6. 6

    Inpaint or upscale as needed

    Separate pass
    Regenerate just the region that is wrong rather than the whole frame. Upscale only if the output is going to print.

The arithmetic on this workflow

Six iterations plus a final render, done properly, costs 18 credits. Six iterations directly on the top model costs 36. Same result, roughly half the allowance.

Reference

Anchoring with a reference image

Dreamina reference image panel with an uploaded image and accompanying prompt
A reference image alongside its prompt. The reference constrains composition and identity; the prompt still controls everything else.

A reference is the answer to the most common complaint about image generators — that the same prompt gives you a different person every time. The model treats the upload as an anchor rather than something to copy, so you keep the face or the product while changing the setting, lighting and framing around it. For any series of images that need to look related, this is not an optimisation but a requirement.

After

Editing what you generated

Dreamina editing canvas showing inpainting and retouch controls
The editing canvas. Inpainting regenerates a masked region only — the cheapest way to fix one wrong detail in an otherwise good render.

Generating again because a hand is wrong is the most common way to waste credits. Masking the hand and regenerating that region alone preserves everything that was already correct and costs a fraction of a full render. Upscaling is worth doing last and only when the output is genuinely going to print — screen use does not need it.

FAQ

Image generation questions

Which Dreamina image model is best?
Seedream 4.0 produces the strongest output and costs the most at 6 credits. But "best" depends on stage rather than quality — Image 3.0 at 2 credits is the right model for the first several attempts, because you are testing composition, not finish.
How many images can I make per day for free?
Between 37 and 112 depending on the model — the daily 225 credits go roughly three times further on the cheapest model than the most expensive.
Can I edit an image after generating it?
Yes. Inpainting lets you regenerate a selected region rather than the whole frame, which is both cheaper and better than re-rolling a full render for one bad detail. Upscaling and retouching are separate passes with their own credit costs.
How do reference images work?
You upload an image and the model uses it as a compositional or stylistic anchor rather than copying it. This is the only dependable way to keep a character or product consistent across several renders — prompt text alone will not reproduce the same face twice.
What resolution do generated images come out at?
Standard renders are suitable for screen use. For print you need an upscaling pass, which is a separate credit cost — worth planning for rather than discovering after you have already spent your allowance on the base render.
Why does my image have garbled text in it?
Text rendering is the weakest area of every current image model. Short words sometimes work; anything longer degrades. Generate the image without text and add real typography in an editor afterwards.

Next

Keep reading

Prompt guide

The formula that gets composition right in fewer attempts.

Aspect ratio picker

Set this before generating, not after.

Video generation

Turning a finished still into motion.

Free credits

How far the daily allowance goes on each model.