Seedream image generator: 4.5, 5.0 Lite and 5.0 Pro in one prompt box

Seedream appears six times in the model list, not three. Each version ships as a pair: one id that works from text alone, one that also takes reference pictures. 4.5 and its editing twin are both 17 credits, 5.0 Lite and its twin 14, 5.0 Pro and its twin 18 at 1K. Price follows the version, not the direction.

All six are switched on today, and they share a dropdown and a credit balance with every other image model here. Running one prompt through 5.0 Pro and then through a rival row is two charges, which still beats guessing.

The dropdown puts 5.0 Pro on top and 4.5 last. That is shelf position, not a verdict.

Four credits separate Lite from Pro, and that is the least interesting part of the choice. The two do not start from the same output size.

Pro is the only Seedream here with a size control. 1K is the base and costs the flat 18; 2K carries a multiplier of exactly 2, so it bills 36. Lite has no size control, and nothing in the form moves its 14.

Now the part that catches people out. Both render at their own base tier, and base is not the same size on both: on Lite it is a 2K render, on Pro a 1K one. Spending 18 instead of 14 buys a different model at a smaller size. Pro never out-sizes Lite at all: its top setting, 2K, is the size Lite already starts at, and getting there costs 36.

So draft on Lite. Four Lite images cost 56 credits, under two Pro renders at 2K. When a prompt finally lands, re-run it on Pro at 2K and pay the 36 once. Credit packs and plans are priced on their own page; the model does not care which one your balance came from.

Five aspect ratios are open to every Seedream id: 1:1, 16:9, 9:16, 4:3 and 3:4. No per-version gaps, no exceptions.

Resolution is where it goes uneven. Only 5.0 Pro shows 1K and 2K. On 4.5 and 5.0 Lite the control is pinned to one value that never reaches the model, because those two APIs have no resolution field.

Those tiers are real: 4.5 defines two, 5.0 Lite three. None are for sale here, by decision rather than oversight. A tier can only be offered once the pricing table knows what to charge for it, and neither Seedream row carries that entry. GPT Image 1.5 is the contrast: its high tier is priced at five and a half times base, so that model does get a tier selector.

One more line nobody reads until the balance drops: output count. A submission can ask for 1, 2 or 4 images, each a full generation at full price. Four Lite images is 56 credits, not 14.

Every Seedream image-to-image id takes up to ten references per request. Generous, and a quiet trap: ten weak references rarely beat two deliberate ones.

The slots earn their keep on three jobs: the subject to preserve, the setting to move it into, and a frame that fixes the palette. Load the subject first, then name the files in the prompt. Nothing else tells the model that image four is a colour reference.

Editing costs the same as generating on every version, so there is no pricing reason to skip a reference. The public gallery is the fastest way to see which prompts survive contact with a reference image.

4.5's editing id stays enabled at 17 credits on purpose. A prompt tuned against 4.5 does not simply sharpen on 5.0; it renders a different picture. Run your old prompts where they were written.

Asking whether Seedream beats Nano Banana has no general answer. The two fail in different directions, and which failure you can tolerate is a property of your prompt. What the table does answer is price, and how steep the climb to 2K is.

5.0 Pro carries the steepest size multiplier of any image model here: double the resolution, double the bill. Nano Banana 2 charges 1.5x for the same step, GPT Image 2 roughly 1.67x, and Nano Banana Pro nothing at all, its 45-credit floor being its own kind of expensive.

At 1K, Pro's 18 sits mid-table, under Nano Banana 2's 20 and far under Nano Banana Pro's 45. At 2K the order flips: 36 puts Pro above Nano Banana 2 at 30 and GPT Image 2 at 26. If most of your work is 2K, that flip is the decision.