Every so often an AI image release actually changes how you work instead of just making prettier pictures. On October 1, Black Forest Labs launched FLUX 3 Image, and the headline feature is one small business owners will understand immediately: you can now tell the model where things go in the frame, with bounding boxes, and it obeys. Place a logo here, a product there, a headline space up top — coordinates, not hope.
After 50 years of laying out ads the old way — grids, guides, and a lot of respectful begging — I can tell you why this matters. The complaint I hear most about AI images isn’t quality anymore. It’s control. You get a lovely picture with the product in the wrong third of the frame and text crammed where your headline belongs. FLUX 3 Image attacks exactly that problem, and paired with what’s already free in ComfyUI, it changes the math for small budgets. Here’s what it does, what it costs, and what I’d actually run today.
What Bounding Boxes Actually Do
Most image models treat your prompt as a suggestion. FLUX 3 Image treats part of it as a specification. According to the official BFL documentation, you append a JSON list of boxes to the end of your prompt — each box is a region on a grid running from 0 to 1000 on both axes, with a short description of what belongs inside it. The model then places each element in its box.
The Second source confirmation describes the same system: draw a box for every element, describe it, summarize the scene in one line, and the model composes to those coordinates. Want a product photo with the bottle in the lower-left, negative space for your price tag in the upper-right, and a lifestyle figure at the right edge? You write the boxes, not a paragraph of begging.
This is the first time a frontier image model has offered layout control at this level — up to 4K native output and up to 10 reference images in the same generation. Bounding boxes are genuinely new; 4K and multi-reference are catching up to what competitors have shipped this fall, but the combination is what makes it production-useful.
Multi-Turn Editing That Doesn’t Wreck Your Image
The second feature is the one your workflow will feel daily: multi-turn edits that leave everything else alone. Recolor, move, resize, or remove several elements in one request, and — per TechTimes’ technical write-up — pixels outside the selected region are preserved with bit-identical numerical fidelity. Not “mostly the same.” Bit-identical.
If you’ve ever asked an AI editor to swap a background and watched it quietly regenerate the subject’s hands, you know why this matters. For a small business, it means you can iterate a marketing image over a whole afternoon — change the background Tuesday, swap the product angle Wednesday, update the accent color Friday — and each edit builds on a stable base instead of re-rolling the dice.
The catch, for now, is access. FLUX 3 Image runs through Black Forest Labs’ API and playground. There is no open-weight download yet — BFL has announced that open weights are “coming in the coming weeks” with no date attached, and community guides are blunt that ComfyUI cannot run FLUX 3 until those weights exist. Any “FLUX 3 checkpoint” you find on a model site today is either a renamed FLUX.2 file or bait. So let’s talk about what runs free today.
What It Costs — and What’s Free Right Now
The pricing is honest usage-based API billing, not a subscription:
| Resolution tier | Price per image |
|---|---|
| Small (768 sq) | $0.041 |
| Mid tiers | $0.041–$0.607 (by resolution) |
| 4K max | $0.607 |
| Launch discount (through Oct 8) | 50% off all tiers |
At full price, a small web banner costs about four cents. A 4K print hero costs about sixty cents. If you generate fifty iterations to nail a hero image, you’ve spent a few dollars — less than a single month of most subscription tools, and you only pay when you work.
But the budget-conscious move is to know what’s free today:
| Tool | Cost | What it’s best at |
|---|---|---|
| FLUX 3 Image (API) | $0.041–$0.607/image | Pixel-exact layout, multi-turn editing |
| FLUX.2 Klein 4B (ComfyUI) | Free — Apache 2.0 license | Commercial use, modest GPUs (~9 GB VRAM) |
| Qwen-Image-2.1 (ComfyUI) | Free — non-commercial license | RGBA transparency, multi-reference editing |
| Z-Image Turbo (ComfyUI) | Free | Fast drafts, brand-consistent social imagery |
| Midjourney | $10–30/mo | Aesthetics, concept exploration |
| Canva Pro | $15/mo | Templates, speed, non-designers |
Here’s my honest read, the same read I’d give you across the desk. FLUX.2 Klein 4B runs natively in ComfyUI under an Apache 2.0 license — free for commercial use, and it’s genuinely good. Qwen-Image-2.1 and Z-Image Turbo cover transparency work and fast drafts at zero cost. None of them have bounding-box layout control yet. So the practical split for a small business in October 2026: do your exploratory and everyday work free in ComfyUI, and pay pennies on the FLUX 3 API only when you need a layout locked to the pixel — a hero banner, a product composite, anything headed to print. When the FLUX 3 open weights land, that paid step likely collapses into the free stack too. Watch for it, but don’t believe a release date until the weights are on Hugging Face.
A Real Workflow: Product Hero in Four Passes
Let me make this concrete. Suppose you sell hand-made pottery and need a hero image for your homepage. Here’s the workflow I’d run — and one I’d run for you, for the record, because this is exactly the kind of production work a studio earns its keep on.
Pass 1 — Free exploration. Open ComfyUI with Z-Image Turbo or FLUX.2 Klein and generate a dozen loose concepts: “hand-thrown stoneware vase on a linen tablecloth, morning window light, shallow depth of field.” No layout discipline yet — you’re hunting for a direction.
Pass 2 — Lock the layout. Take the winner’s direction to FLUX 3 Image and write boxes: vase centered at coordinates that leave clean space top-right for your headline block, a second vessel at the lower-left for balance, a soft shadow field behind. The model composes to your grid instead of its own taste.
Pass 3 — Iterate without wrecking it. Multi-turn edit: warm the background light, then shift the secondary vessel two inches left, then deepen the shadow. Each pass leaves the vase untouched — no re-rolls, no mutated handles.
Pass 4 — Finish where AI is weakest. Bring the result into your editor, drop in real typography, and check the image at phone width. Total spend: a handful of API cents plus an hour of free local generation. That’s the whole cost stack.
The pattern generalizes: free tools for volume and exploration, precision tools for the final composition. That’s been the shape of professional production for fifty years — cheap roughs, expensive finishes — and AI hasn’t changed it, just repriced it.
The Bigger Picture: Control Beats Novelty
Two studies landed the same week as FLUX 3 and they frame why layout control matters more than another quality bump. A Bain study published October 1 surveyed 1,397 senior marketing executives and found only 6% say AI has significantly impacted performance, despite 47% calling AI a core capability. Kantar’s Media Reactions 2026, out October 2, found 75% of marketers plan to increase AI investment while only 23% of consumers trust AI recommendations.
Read those together and the lesson is uncomfortable but useful: the bottleneck isn’t access to AI, it’s the craft around it. Brands that win with AI share three traits per Bain — centralized strategy, rebuilt workflows, and focus on customer value. Generic AI output is what consumers don’t trust, and nothing says generic like a stock-pose image with your text floating in the wrong place. Pixel-level control is how a small business makes AI output look intentional — on-brand, on-grid, and consistent enough that customers stop noticing the tool and notice the brand.
Questions I Expect You’ll Ask
Can I run FLUX 3 Image in ComfyUI today? No. ComfyUI can only host models with published weights, and FLUX 3’s aren’t out yet — open weights are announced but undated. FLUX.2 Klein 4B and Qwen-Image-2.1 run natively in ComfyUI today and are the free options I’d use.
Is FLUX 3 Image free? Not the new model. It’s API-only at $0.041–$0.607 per image depending on resolution, with a 50% launch discount through October 8. The older FLUX.2 Klein 4B is free under Apache 2.0, including commercial use.
Do I need to be a developer to use bounding boxes? The API takes JSON coordinates, so there’s a technical hurdle. In practice, a designer or studio sets up the workflow once — a template of boxes for your brand layouts — and reuses it. That’s a conversation worth having if the coordinate format makes your eyes glaze.
What should a small business actually do this month? Keep your free ComfyUI stack running for volume work, try the FLUX 3 playground while the discount window is open, and don’t buy any “FLUX 3 checkpoint” downloads — the real open weights haven’t shipped.
Where This Leaves You
The tools keep moving, but the decision framework doesn’t: free and open for exploration and volume, paid precision where the layout has to be exact, and a human eye — yours or a studio’s — deciding what “exact” means for your brand. The businesses getting real results from AI aren’t the ones chasing every model release; they’re the ones with a workflow.
