Seedream 5 Lite: API, Pricing and Code Examples (2026)

Seedream 5 Lite is ByteDance’s fast, lower-cost image model from the Seedream 5 family. It turns a text prompt, or a prompt plus up to 14 reference images, into a 2K or 3K picture. This guide covers what the model does well, how it compares with Seedream 5 Pro and other image models, the cost per image in credits and dollars, and a complete working example you can run today.

What is Seedream 5 Lite?

ByteDance released Seedream 5 Lite on 25 February 2026. It is a text-to-image model, meaning it takes a written description (the prompt) and produces an image that matches it. It also accepts reference images, so you can use it for editing and for keeping a product, character or style consistent across a set of pictures.

The “Lite” label is about speed and cost, not about small output. The model renders at 2K or 3K, which is enough for web hero banners, print-quality social posts and product listings without an upscaling step. (“2K” and “3K” refer roughly to the pixel count of the long edge, about 2,000 and 3,000 pixels.)

Here is what you get when you call it through Unicorn API:

• Model id: seedream-5-lite
• Output: 2K or 3K, selected with the size field
• Reference images: up to 14, passed in ref_images
• Aspect ratios: 16:9, 1:1, 9:16, 21:9, 4:3, 3:2, 2:3
• Batching: the n field returns several images from one request
• Other inputs: prompt, aspect_ratio, output_format
• Price: 5.5 credits per image

Write your prompt like an art brief: subject, setting, lighting, composition, and any text that must appear in the image. Seedream models respond well to concrete nouns and explicit layout instructions (“headline at the top, product centred, white background”) rather than strings of style adjectives.

Seedream 5 Lite vs Seedream 5 Pro

ByteDance followed up on 8 July 2026 with Seedream 5 Pro, the top Seedream tier, aimed at polished illustration and product shots. The two models overlap a lot, so the choice comes down to what you are making and how many images you need.

	Seedream 5 Lite	Seedream 5 Pro
Released	25 Feb 2026	8 Jul 2026
Resolution	2K or 3K	1K or 2K
Reference images	Up to 14	Up to 10
Price per image	5.5 credits	7 credits (1K), 14 credits (2K)
Aspect ratios	16:9, 1:1, 9:16, 21:9, 4:3, 3:2, 2:3	Same, plus 3:4
Best for	Volume, drafts, everyday content, multi-reference work	Final hero images, illustration, product shots where finish matters

A few things stand out. Lite actually goes to a higher resolution than Pro, and it accepts more reference images. Pro’s advantage is finish: cleaner detail and more refined composition on the shots where a client will look closely. At 2K, Pro costs about 2.5x what Lite does per image.

A practical workflow many teams settle on: explore with Seedream 5 Lite, where twenty variations are cheap, then rerun the two or three winning prompts on Pro for the final versions. Because both models share the same request shape, switching is a one-word change in your code.

There is also a third member of the family, Seedream 5 Pro Layer Decomposition. As the name says, it takes an existing image (via image_url) and a prompt and works on separating it into layers. It costs 7 credits at 1K or 1.5K and 14 at 2K. It is a specialist tool for designers who need editable parts, not a replacement for either generator.

Seedream 5 Lite pricing in credits and dollars

Unicorn API prices every model in credits. Monthly plans give you credits at $0.01 each (Starter $25, Creator $50, Studio $100, Pro $200 per month). Top-up credits cost $0.015 each and never expire. Full details are on the pricing page.

Seedream 5 Lite costs 5.5 credits per output image. The total is multiplied by n when you request several images, then rounded up to whole credits for the request. That rounding matters more than it looks:

Request	Credits	Cost on a plan	Cost with top-ups
1 image (n=1)	6	$0.06	$0.09
2 images (n=2)	11	$0.11	$0.165
4 images (n=4)	22	$0.22	$0.33

Requesting images in even-numbered batches avoids paying for a rounded-up half credit on every call. If you generate one image at a time, you pay 6 credits each; in pairs, you pay 5.5 each.

What that means per plan, as a worked example:

• Starter ($25, 2,500 credits): about 416 single-image requests, or about 454 images if you batch in pairs.
• Creator ($50, 5,000 credits): about 833 single images, or about 908 in pairs.
• Studio ($100, 10,000 credits): about 1,666 single images, or about 1,818 in pairs.

You never have to guess. Send the same body you would send to the job endpoint to POST /v1/models/price and you get the exact cost before anything runs. This is worth wiring into any app where users choose the batch size.

A note on Seedream 5 pricing elsewhere: other hosts such as Replicate and fal also serve Seedream models, each with its own account, billing and request format. The reason to use Unicorn API is that the same key and the same credit balance also cover video, audio and chat models, so you are not juggling a separate bill for every provider.

How to use the Seedream 5 Lite API

Image generation is asynchronous. You submit a job, get back a job id, and poll until the job finishes. The flow is the same for every media model on Unicorn API, which makes it easy to swap models later.

1. Get an API key

Create a key on the API keys page and export it in your shell:

export UNICORN_API_KEY="your-key-here"

If you would rather test prompts before writing code, the image playground runs the same models in the browser.

2. Check the price

curl -X POST https://api.unicornapi.net/v1/models/price \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedream-5-lite",
    "input": {
      "prompt": "Minimalist poster for a coffee brand, the word BREW in bold sans-serif at the top, a single ceramic cup centred on a warm beige background",
      "aspect_ratio": "2:3",
      "n": 2
    }
  }'

3. Submit the job and poll for the result

Here is a complete Python script. It creates the job, polls every few seconds, stops on failure, and prints the URL of each image.

import os
import time
import requests

BASE_URL = "https://api.unicornapi.net"
HEADERS = {
    "Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}",
    "Content-Type": "application/json",
}

body = {
    "model": "seedream-5-lite",
    "input": {
        "prompt": (
            "Minimalist poster for a coffee brand, the word BREW in bold "
            "sans-serif at the top, a single ceramic cup centred on a warm "
            "beige background, soft window light from the left"
        ),
        "aspect_ratio": "2:3",
        "n": 2,
    },
}

# Create the job
resp = requests.post(f"{BASE_URL}/v1/jobs", headers=HEADERS, json=body)
resp.raise_for_status()
job = resp.json()
job_id = job["id"]
print("Job created:", job_id)

# Poll until it finishes
while True:
    resp = requests.get(f"{BASE_URL}/v1/jobs/{job_id}", headers=HEADERS)
    resp.raise_for_status()
    job = resp.json()
    status = job["status"]
    if status == "succeeded":
        break
    if status in ("failed", "canceled"):
        raise RuntimeError(f"Job {job_id} ended with status {status}: {job}")
    time.sleep(3)

for i, output in enumerate(job["outputs"]):
    print(f"Image {i + 1}: {output['url']}")

With a single image, job["outputs"][0]["url"] is all you need. Download the file promptly and store it in your own bucket rather than hotlinking the output URL from your production site.

The same two calls work from curl if you just want to try it:

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"seedream-5-lite","input":{"prompt":"A red fox resting in a snowy forest at dawn, soft light, highly detailed","aspect_ratio":"16:9"}}'

# then, using the id from the response:
curl https://api.unicornapi.net/v1/jobs/JOB_ID \
  -H "Authorization: Bearer $UNICORN_API_KEY"

For the full list of fields, including the accepted values for size and output_format, see the API documentation. If you work inside Cursor, Claude Code or another coding agent, you can also connect the API over MCP and ask the agent to generate images directly.

Editing and consistency with reference images

The 14-image reference limit is the feature that sets Seedream 5 Lite apart from most budget image models. A reference image is an existing picture you hand the model so it can copy identity, style, layout or a product’s exact look. More references give the model more angles and more context.

Typical uses:

• Product variations: send three or four photos of the same sneaker from different angles and ask for it on a beach, in a studio, on a city street.
• Character consistency: send several images of a mascot and ask for new poses and scenes, so a comic or campaign keeps the same face.
• Style transfer: mix two or three style references with a subject reference and describe how they combine.
• Edits: send one image and describe the change: “same room, replace the sofa with a green velvet one, keep the lighting”.

The request is the same job call with a ref_images array of publicly reachable image URLs:

body = {
    "model": "seedream-5-lite",
    "input": {
        "prompt": (
            "The sneaker from the reference photos on wet city pavement at "
            "night, neon reflections, low camera angle, product in sharp focus"
        ),
        "ref_images": [
            "https://example.com/sneaker-side.png",
            "https://example.com/sneaker-front.png",
            "https://example.com/sneaker-sole.png",
        ],
        "aspect_ratio": "4:3",
    },
}

Tell the model what each reference is for. “Use the first image for the shoe, the second for the colour palette” works much better than a list of images and a vague prompt. Without guidance, the model has to guess which details to keep.

How Seedream 5 Lite compares with other image models

Seedream 5 Lite sits in the middle of the price range. It is not the cheapest image model available, but it is one of the few low-cost options with 2K to 3K output and a large reference limit. Here is how it compares with other models on the same key (prices are per image before rounding; every model rounds up to whole credits per request):

Model	Max resolution	Reference images	Credits per image	Good at
Seedream 5 Lite	3K	Up to 14	5.5	High-res everyday work, multi-reference consistency
Seedream 5 Pro	2K	Up to 10	7 (1K), 14 (2K)	Polished illustration and product shots
Nano Banana 2 Lite	1K	Up to 10	4	Cheapest Gemini option, fast 1K images and edits
GPT Image 1.5	Quality tiers	Up to 4	3 (low), 4 (medium), 22 (high)	Prompt adherence, legible text, transparent backgrounds
Flux 3 Klein Text to Image	4K	None (text only)	1.6 (1K), 2.5 (2K), 6.5 (4K)	Very cheap drafts, 4K output
Flux 3 Klein Edit Image	4K	Input images via image_urls	1.6 (1K), 2.5 (2K), 6.5 (4K)	Low-cost editing of existing images

How to choose:

• Pick Seedream 5 Lite when you need sharp, large images in volume, or when you have many reference photos of one product or character.
• Pick Seedream 5 Pro for the final image a client or customer will scrutinise.
• Pick GPT Image 1.5 when the image needs a transparent background, for stickers or logos dropped onto other designs.
• Pick Nano Banana 2 Lite or Flux 3 Klein when you only need 1K drafts and price is the deciding factor. (If you are weighing “free” Gemini image offers, our look at Nano Banana’s free unlimited claim covers the real limits.)

Because all of these use the same job endpoint, the cleanest approach is to make the model id a config value. Run the same ten prompts through two or three candidates, compare the outputs side by side, and decide with your own images instead of someone else’s benchmark. The full list is in the model catalog.

Common mistakes with Seedream 5 Lite

1. Requesting single images in a loop. Each n=1 request rounds 5.5 credits up to 6. If you need many images from one prompt, set n to an even number.
2. Using an aspect ratio the model does not offer. Seedream 5 Lite supports 16:9, 1:1, 9:16, 21:9, 4:3, 3:2 and 2:3. Portrait 3:4 is a Seedream 5 Pro option. If you need 3:4 from Lite, generate at 2:3 or 1:1 and crop.
3. Passing private image URLs as references. The model has to download each reference. Links behind a login, expiring signed URLs with very short lifetimes, or localhost addresses fail. Use public or long-lived signed URLs.
4. Stacking 14 references with no instructions. More references only help if the prompt says what to take from each. Conflicting references produce muddled results.
5. Polling too aggressively. Checking every 100 milliseconds wastes requests. A three to five second interval is plenty for image jobs.
6. Not handling failure states. Treat any non-success final status as an error and log the job body, so you can see why a prompt was rejected.
7. Paying Pro prices for drafts. At 2K, Seedream 5 Pro is about 2.5x the cost of Lite. Use Lite to explore and Pro for finals.

Is there a free Seedream 5 Lite app or APK?

Searches for “Seedream 5 Lite mod apk” and “Seedream 5 Lite free download” are common, so it is worth being direct. Seedream 5 Lite is a hosted model: it runs on remote GPU servers, and every app that offers it is sending your prompt to a server and returning the result. There is no file you install that puts the model on your phone.

That means an APK advertised as a “mod” or “unlocked” version cannot contain the model itself. At best it is a wrapper around someone else’s account; at worst it is malware. Treat these packages as untrusted and keep them off devices that hold work or banking data.

Legitimate free access usually means a consumer app with a small daily allowance, watermarks or limited resolution. That is fine for trying the model. For anything you build, ship or sell, an API key gives you predictable pricing, full 2K or 3K output, and code you control. On Unicorn API, Seedream 5 Lite works out to about 6 cents for a single image on a monthly plan, and the price endpoint tells you the exact cost of every request before it runs.

If you are ready to try it, create a key, paste the Python script above, and change the prompt to something you actually need this week. The fastest way to judge any image model is to run it on your own work.