
FLUX 3 Image Release Date: Launch, Access, and Weights
FLUX 3 Image launched October 1, 2026. See how its release differs from the July family announcement, the Video launch, and public model-weight availability.
FLUX 3 Image was released on October 1, 2026. Black Forest Labs places its Image release under that date in the official release notes, with generation, image editing, reference images, and a dedicated Image endpoint. That is the direct source for the date, rather than an inference from a promotion or a social post.[1]
The earlier July 23 announcement introduced the wider FLUX 3 family and said Image early access would follow. It did not describe the Image product as publicly released that day. This distinction matters if you are deciding whether to keep waiting, which documentation to use, or whether a download named FLUX 3 contains the image model you want.[2]
TL;DR
- October 1, 2026 is the documented FLUX 3 Image release date. The official entry names the Image product and links its endpoint and guides.[1]
- July 23 was the family announcement. Its launch plan separated Video, Action, Image, and the future Dev open-weight backbone.[2]
- August 4 was the Video preview release. Its endpoint generates video with synchronized audio; it is a different release from Image.[1]
- Image has current API documentation. BFL documents text generation, editing, and up to ten references through
flux-3-image.[3] - A commercial weights offer is separate from public downloads. The Image product page offers licensing for companies that want to fine-tune and deploy on their own infrastructure.[4]
Three dates explain the FLUX 3 Image release
The rollout is easier to understand when every date carries the name of the product or milestone it actually describes.
| Date | What BFL announced | What the date establishes |
|---|---|---|
| July 23, 2026 | FLUX 3 multimodal family | The family announcement and a phased rollout plan |
| August 4, 2026 | FLUX 3 Video preview | A documented video endpoint with synchronized audio |
| October 1, 2026 | FLUX 3 Image | The image generation and editing release |
The first date comes from the original family announcement; the latter two have their own entries in BFL's official release notes.[2][1]
The July article already showed image capabilities, so someone who saw those examples can reasonably remember FLUX 3 as a summer release. But the Image section described preliminary results and a forthcoming early-access phase. Its launch plan treated image synthesis and editing through APIs and private weights as one stage, and a later open-weight backbone as another.[2]
An announced capability, an early-access invitation, and a documented product launch are useful milestones. They answer different questions. For “when did FLUX 3 Image launch?”, use October 1. For the history of the multimodal model family, July 23 remains the relevant starting point.
The October release note supplies a calendar date. It does not supply a launch hour or a timezone, so this guide does not attach a precise timestamp to it.
What arrived with the Image release
The October entry describes one endpoint for generating a new image, editing an existing image, and combining references. It highlights prompt-based bounding boxes, local edits, up to ten reference images, output up to 4K, and fifteen explicit aspect ratios. The prompt determines the operation; BFL says there is no separate mode field.[1]
Those details distinguish the Image product from a FLUX 3 video request that happens to begin with a still picture. Image-to-video uses an image as input to produce motion. The Image endpoint produces a still image, including an edited or newly composed one.[1][3]
The current overview lists five output tiers: 768sq, 1k, 1.5k, 2k, and 4k, with 1k as the default. It also documents grounding, enabled by default, which allows web and image searches before generation. These are current documented controls, not promises that any particular image will satisfy a brief.[3]
Use the current overview for request options when integrating. A release note is a dated account of a launch; it is not necessarily the most complete parameter reference after the product evolves. That matters here because the current parameter table includes the intermediate 1.5k tier.
This article establishes when the product was released and which access paths BFL describes. It does not report an image-quality benchmark or an end-to-end latency test. For practical prompt structure, use the separate FLUX 3 Image editing guide.
Where you can access FLUX 3 Image
As checked on October 9, BFL's product page links a browser tool and its documentation. The overview documents POST https://api.bfl.ai/v1/flux-3-image and an asynchronous result workflow. The same product page invites companies to discuss commercial weights for their own infrastructure.[4][3]
Choose the route that matches the work:
| Your immediate task | Relevant access path | What to establish before relying on it |
|---|---|---|
| Explore the model interactively | BFL's linked browser tool | Available controls and the account requirements shown there |
| Add generation or editing to an application | The documented Image endpoint | Authentication, input validation, task polling, and output handling |
| Operate the model on your own infrastructure | The commercial weights discussion | Exact model package, contract scope, deployment requirements, and support |
The last column is a suggested evaluation procedure, not a claim that every account has identical access or that a private license has already been granted.
For the reAPI interface, the FLUX 3 Image model page and request documentation describe its controls, pricing, and task format. Use the documentation for the service you are actually calling. BFL's direct request names and authentication should not be pasted unchanged into a different service's endpoint.
The public documentation is evidence that the product has a described access path. A production integration still needs its own request and output checks. Neither a launch announcement nor a successfully opened documentation page is a substitute for verifying your application's behavior.
A product launch does not settle the weights question
The FLUX 3 Image product page makes a specific current offer: a commercial weights license for companies running generation at scale, including fine-tuning and deployment on their infrastructure. It does not present that offer as a public checkpoint download.[4]
On October 9, we checked both pages of Black Forest Labs' public Hugging Face model catalog and its complete public GitHub repository list. We found no FLUX 3 Image or FLUX 3 Dev checkpoint in those catalogs. This is a dated observation about the official public listings, not a claim about private deliveries to commercial customers.[5][6]
That distinction prevents two opposite mistakes. A missing public download does not cancel the documented hosted release. A hosted release also does not mean a downloadable package with the same capabilities, license, and deployment instructions has appeared.
The July roadmap named FLUX 3 Dev as the planned open-weight multimodal backbone. It was separate from the Image API and private-weight stage. A future-oriented statement from that roadmap should remain future-oriented unless a later checkpoint or release establishes otherwise.[2]
If local execution is your requirement, the FLUX 3 Image weights and open-source guide covers the catalog evidence and the questions to resolve with BFL. It also explains why a FLUX 3 Action download or an open-source API client is not the Image checkpoint.
Read a release claim against the exact product
When a page says “FLUX 3 is available,” first identify the noun behind the claim. Does its request create a video, a still image, or robot actions? Does the link lead to a hosted endpoint, a model card with files, or source code that calls a remote service?
For FLUX 3 Image, a useful reference record contains the product name, the October 1 release-note link, the current Image overview, and the date you checked access. Keep the July family announcement as history. This lets someone reviewing the record follow the same evidence without guessing which release you meant.
Be equally specific about your own application. “The model was released” describes BFL's milestone. “Our integration accepts this request” requires evidence from your integration. “This workflow preserves our product packaging” needs an output review against that packaging. Keeping those statements separate makes the release date useful without making it do more work than it can support.
Our recommendation is to start evaluation from the simplest task that matters to your project. A catalog team might check whether an edit retains a particular product shape; an editorial team might check its illustration brief. Those are example acceptance questions, not reported tests or proof that one model is best for every team.
Questions about the FLUX 3 Image launch
What is the FLUX 3 Image release date?
October 1, 2026. BFL's official release notes place an entry specifically titled FLUX 3 Image under that date and link the Image endpoint and guides. That is the strongest direct source for the public release date used here.[1]
Why do some pages say July 23, 2026?
July 23 is the date of the wider FLUX 3 announcement. The original post discussed Image capabilities but said its early-access phase would open in the following weeks. It also laid out a staged rollout, so that date should not be substituted for the October Image release.[2]
Was the August 4 launch the same product?
No. The August 4 release-note entry concerns FLUX 3 Video with synchronized audio, available as a preview, and names the Video endpoint. Its still-image inputs and Image-to-Video mode do not make it the FLUX 3 Image product.[1]
Is FLUX 3 Image still just a waitlist announcement?
BFL now publishes an Image endpoint reference and an Image product page with a browser-tool link. Those sources establish more than the July invitation to early access. This guide checked the documentation on October 9; it did not test every account's entitlement or make a paid generation.[3][4]
Did the October launch include public downloadable weights?
The dated Image entry documents the generation and editing endpoint. Separately, BFL's product page offers commercial weights. We found no Image or Dev checkpoint in the official public catalogs checked on October 9, so the hosted release should not be described as a verified public-weight release.[1][4][5]
Should I use launch-day examples or current documentation?
Use the release note to establish the historical date, then use the current documentation for request fields and limits. For FLUX 3 Image, the current overview supplies the resolution choices, reference limit, grounding control, and polling behavior needed to evaluate the documented interface.[3]
Start from the current Image entry
Use October 1, 2026 when recording the FLUX 3 Image launch, then attach the current request documentation to your evaluation. Choose the hosted workflow if that meets your needs; investigate the commercial weights offer if your requirement is local deployment. Keep your own integration results alongside those sources so the next person can distinguish the product's release from what your application has actually verified.
References
- Black Forest Labs. Release Notes: October 1, August 4, and July 23, 2026. Retrieved October 9, 2026 from docs.bfl.ai/release-notes.
- Black Forest Labs. FLUX 3 — Real World Models. July 23, 2026; retrieved October 9, 2026 from bfl.ai/blog/flux-3.
- Black Forest Labs. FLUX 3 Image overview. Retrieved October 9, 2026 from docs.bfl.ai/flux_3/flux3_image_overview.
- Black Forest Labs. FLUX 3 Image product page. Retrieved October 9, 2026 from bfl.ai/models/flux-3-image.
- Black Forest Labs on Hugging Face. Official model catalog, both pages. Observed October 9, 2026 at huggingface.co/black-forest-labs/models and page 2.
- Black Forest Labs on GitHub. Official public repository list. Observed October 9, 2026 at github.com/black-forest-labs.
Author

Categories
More Posts

AI Image Generation Cost vs Video: 2-9x More Per Frame
AI image generation cost per megapixel runs 2 to 9 times a frame of generated video. The per-frame math across eight video tiers and seven image models.


How to Control AI Video Camera Movement with a Phone Reference
Record natural camera movement with a phone, turn it into a clean motion reference, and guide an AI video model without learning 3D animation first.


MAI Image 2.6 pricing: estimates, references, and Flash
Understand MAI Image 2.6 pricing, why final image costs can differ from estimates, how references affect the bill, and what Microsoft says about Flash.
