Choosing among the FLUX models in the catalog

Updated 2026-10-02

FLUX is a family from Black Forest Labs, and the catalog carries more of its members than most people expect: the FLUX.2 line in dev, pro, max, flex and two klein sizes, flux-3-image, the Kontext editing models, and earlier FLUX.1 entries. The naming does not tell you which to pick. This page separates what VideoRouter's documentation and catalog actually say about each group from what you have to measure yourself. It makes no claims about relative image quality, because none are documented.

Get the exact ids first

Slugs vary in case and punctuation across the catalog (for example flux-3-image and flux-kontext-pro versus the FLUX.2 names), so do not type them from memory. List them:

import requests

models = requests.get("https://videorouter.sh/api/v1/images/models",
                      headers={"Authorization": "Bearer llmr_sk_live_..."}).json()["data"]
flux = sorted(m["id"] for m in models if "flux" in m["id"].lower())
print(flux)

GET /v1/images/models is unbilled and returns bare ids; the models index has the per-model pages with hosts and prices.

The groups, and what is known

One important detail: not every FLUX id accepts input_references. Only edit-capable models do. If your product needs editing, shortlist flux-3-image and the Kontext ids first rather than discovering a 400 later.

A decision order that works

1. Do you need image-to-image? If so, restrict to the edit-capable ids above. Everything else is text-to-image only.
2. Do you control output size? Per-megapixel pricing on some hosts means large sizes cost more on those. Flat-priced ids are easier to budget. The model page states the basis.
3. What is the stage? Exploration and thumbnails are where a smaller or cheaper variant earns its place. Delivery assets are where you compare the top ids.
4. Which host? The same id may be sold by several hosts at different prices. Let routing pick the cheapest healthy one unless you have a reason to pin. The model/host suffix expresses a preference and still allows fallback.

An A/B harness

import itertools, json, requests

API = "https://videorouter.sh/api/v1"
H = {"Authorization": "Bearer llmr_sk_live_..."}
SHORTLIST = ["flux-3-image", "flux-kontext-pro"]   # plus the FLUX.2 ids from /images/models
PROMPTS = json.load(open("prompts.json"))          # 20-30 prompts that look like your traffic

rows = []
for model, (i, prompt) in itertools.product(SHORTLIST, enumerate(PROMPTS)):
    r = requests.post(f"{API}/images", headers=H,
                      json={"model": model, "prompt": prompt, "aspect_ratio": "1:1"}).json()
    rows.append({"model": model, "prompt_id": i, "cost": r["usage"]["cost"],
                 "image": r["data"][0].get("b64_json") or r["data"][0].get("url")})
json.dump(rows, open("ab_run.json", "w"))

Keep everything but model constant, including size. Then shuffle and strip model names, have the people who accept the output score each image against a short rubric (prompt adherence, text legibility, subject fidelity, artifacts), and join the scores back to the cost you logged. Compare cost per accepted image, not list price. Kontext and flux-3-image need a second run with input_references on your real edit inputs, scored on how well the original subject is preserved.

Common selection mistakes

• Comparing across sizes. If one candidate is priced per megapixel and another flat, a test at a single small size flatters the first. Test at the sizes you will ship.
• Judging on a handful of prompts. Variance between prompts is large. Use enough prompts that one lucky or unlucky image cannot decide the result.
• Locking in on a slug. Keep the model id in config. A newer FLUX release should be a one-line change followed by a re-run of the harness.
• Ignoring the host spread. The same id can differ in price by host, so a cheaper-looking id may not be cheaper once you compare the same variant across hosts.

Live comparison

Per image, before VideoRouter's 2% platform fee. For tiered models each row compares the resolution tier with the widest host-to-host gap. Built 2026-10-02 from the live catalog.

Use the table to see the cheapest and priciest host for each FLUX id and how wide the gap is. Because it is rebuilt from the catalog, re-check it when you are about to commit to a volume. When you have a winner, keep the model id in config rather than in code, and re-run the harness when Black Forest Labs ships a new version.

See the cross-family comparison for FLUX against other families, the GPT Image parameter guide for the OpenAI-only controls that do not apply to FLUX, the quickstart for a first request, and videorouter.sh/signup for a key.

Frequently asked questions

Keep reading

• Choosing an Image Generation API — Price, Quality and Control
• Text-to-Image vs Image-to-Image API: Modes, Fields, Pitfalls
• GPT Image vs FLUX vs Seedream vs Nano Banana: How to Choose
• Image Generation API Cost Control: Billing, Drafts, Caching