
October 17, 2026 • 7 Min Read
October 26, 2026 • 11 Min Read

Melius fits projects where Flux is one step among several: it runs FLUX 3 Image and FLUX.2 Max beside other image and video models on one canvas. Black Forest Labs' API and Dev Playground are the official source, fal.ai serves both FLUX 3 Image and FLUX 3 Image Klein, Replicate keeps 35 Black Forest Labs models behind one API, Krea runs Flux in a realtime web app, Magnific pairs FLUX.2 with a stock library, Adobe Firefly puts FLUX.2 inside Photoshop, and ComfyUI is where you run the open weights yourself.
We compared the platforms on which current Flux models each one serves, whether you work in a browser or through code, what you can do with an image after it's generated, and whether the license covers commercial work.
Black Forest Labs ships Flux in three generations, and most platforms carry only part of the lineup. FLUX 3 Image reached the API on October 1, 2026, and FLUX 3 Image Klein, a smaller open-weights version, followed on October 15. FLUX.2 is still widely served, and FLUX.1 Kontext remains in the API, although BFL now recommends FLUX.2 for new projects. FLUX 3 also covers video, so "FLUX 3" on its own can mean a video model.
Model | Access | What it's known for |
FLUX 3 Image | API; commercial weights license through BFL sales | Layout control with bounding boxes, up to 10 reference images, output up to 4K |
FLUX 3 Image Klein | Open weights (5.6B), FLUX Kommunity License | The same request format as FLUX 3 Image in a model small enough to self-host |
FLUX.2 Max | API and BFL's playground; no open weights | BFL's top FLUX.2 tier, with grounded web search |
FLUX.2 Pro and Flex | API | Exact hex color control; Flex is tuned for typography and small details |
FLUX.2 Dev | Open weights (32B), non-commercial license | The open FLUX.2 model, hosted by most API platforms |
FLUX.2 Klein | Open weights: 4B under Apache 2.0, 9B non-commercial | Small, fast models for interactive use |
FLUX.1 Kontext | API; Dev weights non-commercial | Instruction-based image editing |
Platform | Current Flux models | How you use it | Best for |
Melius | FLUX 3 Image, FLUX.2 Max, FLUX1.1 Pro Ultra, Flux Schnell, plus the FLUX 3 video models | Browser canvas with the Mel agent, plus MCP, the mel CLI, a REST API and Slack | Projects where Flux is one step among several |
Black Forest Labs | FLUX 3 Image, FLUX 3 Video, FLUX.2 and the FLUX.1 API models | API, Dev Playground, MCP server | Developers who want first-party access |
fal.ai | FLUX 3 Image, FLUX 3 Image Klein, FLUX.2 Max, Flex and Klein, Kontext Pro | API with per-model playgrounds, Sandbox, MCP, CLI | Getting new Flux releases quickly |
Replicate | FLUX 3 Image, FLUX.2 Max, Pro, Flex, Dev and Klein, Kontext, LoRA variants | API | Pipelines that need many Flux variants |
Krea | FLUX 3 Image (early access), FLUX.2 Pro, Flex, Max and Klein, Kontext | No-code web app | Realtime, hands-on image work |
Magnific | FLUX.2 Pro, Max, Flex and Klein, Kontext | No-code web app, plus an API | Designers who also want stock assets |
Adobe Firefly | FLUX.2, FLUX.2 Klein 4B, FLUX1.1, Kontext | Firefly app, Boards, Photoshop, Illustrator | Creative Cloud users |
ComfyUI | Open weights: FLUX 3 Image Klein, FLUX.2 Dev and Klein, FLUX.1 | Local node graph, or Comfy Cloud in the browser | Running open weights with full control |
Melius is the AI-native operating system for creatives, with AI agents for creative work on a node-based canvas. FLUX 3 Image, FLUX.2 Max, FLUX1.1 Pro Ultra and Flux Schnell sit in our library of leading models on one subscription. Brief Mel, the Melius agent, and it builds the nodes and picks a model per step, or you set Flux on a node yourself.
On Melius, Flux can be used alongside GPT Image 2.5 to render legible label and packaging text, and alongside Nano Banana Pro to match a reference look closely. When the still becomes video, it's often paired with Seedance 2.5 to keep a character consistent across clips, or with Veo 3.1 to add cinematic product motion.
What it does well: A FLUX 3 Image still, at up to 4K and edited from as many as 10 references, can feed an upscale, a resize per channel and an image-to-video node on one canvas, and changing an upstream step reruns everything downstream of it. The same canvases run from Claude or ChatGPT desktop over MCP, from the mel CLI or through the REST API, so a Flux step can sit inside a script or another agent's workflow. Professional plans and up add a Slack agent.
Trade-off: Melius carries four Flux image models, plus FLUX.1 and FLUX.2 LoRAs on Enterprise. FLUX.2 Pro, Flex, Dev and Klein and FLUX.1 Kontext aren't in the catalog.
Best for: Teams for whom Flux is one step in a campaign or product shoot, tested against other models without switching tools.
Black Forest Labs makes Flux, and its API is the first-party way to call the current models, from FLUX 3 Image to FLUX.2 and the FLUX.1 Kontext editors. The consumer FLUX Playground closed on August 31, 2026. Its replacement, the Dev Playground in the BFL dashboard, uses the same BFL account for text-to-image and multi-reference editing.
What it does well: Output from the API carries full commercial rights with no separate license. A FLUX MCP server lets you generate from Claude, Cursor and similar tools without writing API code, and BFL hosts free demos of FLUX.2 Klein and its Flux Tools for erasing, outpainting and virtual try-on.
Trade-off: The Dev Playground drops the old Playground's saved styles, community gallery and Fill and Expand tools. Anything beyond generating and editing single images has to be built on the API yourself.
Best for: Developers who want first-party access and the official documentation.
fal.ai is a generative media platform for developers. It serves both FLUX 3 Image and FLUX 3 Image Klein, alongside FLUX.2 Max, Flex and Klein and FLUX.1 Kontext Pro, and it's where Black Forest Labs sends people to try FLUX 3 Max, its fast video model.
What it does well: Every model has a browser playground, and the Sandbox runs one prompt across several models at once. The same models are available through the API, an MCP server and a CLI, and a Flux LoRA trainer handles fine-tuning.
Trade-off: It's built for developers. The playgrounds are built for testing, with nowhere to keep a project together, and pricing for FLUX 3 Image Klein was still listed as planned in October 2026.
Best for: Product teams that want FLUX 3 Image and Klein now, with a playground to test them first.
Replicate runs models behind a single API, and its Black Forest Labs page lists 35 models, among them FLUX 3 Image, FLUX.2 Max, Pro, Flex, Dev and Klein (including base and LoRA versions), the Kontext models and FLUX.1 Krea Dev. Cloudflare acquired Replicate in December 2025, and Replicate said at the announcement that its API wasn't changing.
What it does well: Older FLUX.1 models stay available next to the new ones, so a pipeline built on one version doesn't break when the next ships. One API works the same from Python, Node or plain HTTP, and Cog packages a custom model for deployment.
Trade-off: FLUX 3 Image Klein isn't on Replicate yet, and its web interface is one run form per model.
Best for: Teams whose pipelines depend on many Flux variants and fine-tunes.
Krea is a web app for AI image and video generation. It has FLUX 3 Image in early access, FLUX.2 Pro, Flex, Max and Klein, FLUX1.1 Pro and Ultra, and FLUX.1 Kontext, and it co-developed FLUX.1 Krea Dev with Black Forest Labs.
What it does well: Realtime generation redraws the image as you change the input, an enhancer upscales the result, and LoRA training for images and video is built in, all without code.
Trade-off: FLUX 3 Image is still in early access on Krea, so the newest model isn't a settled part of the product yet.
Best for: Creatives who want hands-on, interactive Flux work in the browser.
Magnific is the name Freepik took in April 2026. Its image generator has FLUX.2 Pro, Max, Flex and Klein plus Flux Kontext among more than 15 image models, and the same subscription includes a stock library of over 250 million assets.
What it does well: Premium+ and Pro include unlimited FLUX.2 Pro generations at 1K, and Pro adds unlimited FLUX.2 Max at 1K, which suits high-volume exploration. Enterprise adds SSO and legal indemnification, and self-serve plans can use the API.
Trade-off: Magnific hadn't added a FLUX 3 image model as of October 2026.
Best for: Designers and marketers who want Flux and stock assets on one plan.
Adobe offers Flux as a partner model. FLUX.2 runs in Firefly's Text to Image, Edit Image and Boards, in Photoshop's Generative Fill and in Illustrator's Concept to Vector, and FLUX1.1 and FLUX.1 Kontext are available in Boards, Text to Image and Photoshop's Generative Fill.
What it does well: Flux output lands in the apps many designers already finish their work in, and Boards puts Flux images on a shared moodboard next to Adobe's own models. Firefly accepts up to four reference images for FLUX.2.
Trade-off: Adobe's commercial-safety position covers its own Firefly models, not partner models, so for Flux output the commercial-use call is the creator's. There's no FLUX 3 in Firefly yet.
Best for: Designers who already pay for Creative Cloud.
ComfyUI is an open-source, node-based interface for running image and video models locally, and Comfy Cloud runs the same graphs in the browser. It's supported FLUX 3 Image Klein since release day and runs the other open Flux weights from Hugging Face, including FLUX.2 Dev and Klein.
What it does well: You control every part of the pipeline, down to the sampler, and the INT8 build of FLUX 3 Image Klein fits on an 8 GB RTX card. Comfy says that every model on Comfy Cloud is cleared for commercial use.
Trade-off: The learning curve is steep, and on a local install the license is yours to check: FLUX.2 Dev and the 9B Klein models are non-commercial. A compromised ComfyUI extension stole user data in 2024, so install custom nodes only from sources you trust.
Best for: Technical users who want open weights and full control.
Start with a brief to Mel that says what the image is for, attaches your references and names Flux where you want it:
Here's our serum bottle and two lifestyle photos for mood. Make a 16:9 hero for the site with FLUX 3 Image and keep the label text exactly as it is. Then run the same prompt through FLUX.2 Max and Nano Banana Pro so I can compare. Upscale whichever I pick and resize it for an Instagram post and a 4:5 ad.
Mel lays out the image, upscale and resize nodes, and asks when the brief leaves something open, such as the background or the audience. From there you can change the model on any node or rewrite a prompt and rerun it. An inpaint fixes one area of the frame and leaves the rest alone. You can also hand the same job to Claude or ChatGPT desktop through the Melius MCP connector, or script it with the mel CLI. Our guide to one subscription for every model covers how the rest of the library fits around Flux.
For other image models worth testing against Flux, see our roundup of the best AI image generators for creators and our breakdown of which Nano Banana model to use.
Want to put FLUX 3 Image up against the rest of the library? Open FLUX 3 Image on Melius and brief Mel with your first shot.

October 17, 2026 • 7 Min Read

October 17, 2026 • 12 Min Read

October 15, 2026 • 9 Min Read