Create composed images from a prompt or references, refine details in Oeneo AI workspace.
BFL official examples
Explore targeted photo edits, multi-reference composition and native 4K image generation. These BFL official examples show the input, the intended change and the details to inspect in the result.
Targeted image editing
Add a print to an existing portrait with an image-editing instruction. In this official example, the cat graphic appears on the white T-shirt while the face, pose and background remain recognizable. Inspect those details after each edit.
Interior design variations
Turn a room photo into an interior-design variation by describing only the furniture change. The foreground chairs become a blue sofa; windows, wall panels and the rug give you a clear before-and-after comparison.
Multi-reference composition
Give each reference a role: the owl, the sparrow and the forest. FLUX 3 Image brings them into one composition with shared lighting and scale. The three source images appear below the official result. You can upload up to 10 references; describe each image's role and check which details carry through.
Native 4K image generation
Generate a storefront scene where signs, materials and small figures matter. BFL's original 5,456 × 3,072 soba-shop image is shown with sign and diner crops underneath. Those are details from the same file, not extra input images or upscaled results. Open the original to inspect the full resolution.
BFL official examples
Choose FLUX 3 Image when the image needs more than a subject: a deliberate layout, a focused edit, readable type or several visual references working together. These are model characteristics, not guarantees of unchanged pixels or perfect spelling.
Describe placement, hierarchy and negative space as part of the prompt. FLUX 3 Image is designed to reason about how elements share a frame, which helps when a poster or editorial layout needs structure before polish.
Name one change and the parts that matter around it. This focused editing approach is useful when an approved subject, pose or room should remain recognizable while one variable changes. Review every round for drift.
Put exact short copy in quotation marks and describe its style and position. The model is promoted for text inside images and multiple languages; proofread names, dates and characters before publication.
Use several references when the result needs a particular subject, material or setting. Assign each input a role, keep the order clear, and treat the result as a new composition rather than an exact product or identity copy.
Describe the result in plain language. For an edit, focus on the change; for a composition, explain how the elements relate.
Place the subject and the text
Specify left or right, foreground or background, and where the title needs clear space. Put exact short text in quotation marks.
Name the change and the parts to preserve
For example: add a blue print to the shirt; keep the face, pose, lighting and background. Avoid redescribing the entire image.
Assign a role to every reference
Identify the subject, clothing, material or setting from each image. Keep upload order consistent with the brief.
Proofread lettering and product details
Readable text and reference guidance are capabilities, not guarantees. Inspect names, dates, faces and packaging before use.
Check references before editing
fal accepts 1–10 images, each at least 256 pixels wide and tall and at most 4 MP. Large generated images may need resizing before another edit.
Bounding boxes are a separate workflow
BFL demonstrates a visual layout canvas. This Oeneo AI page does not offer a bounding-box or mask editor.
Compare the current Oeneo AI input capacity, output choices and credit costs. The prices use each model's default quality, one output and one reference for editing. Changing quality or adding references can change the quote.
| Specification | FLUX 3 Image | Ideogram 4.5 | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst |
|---|---|---|---|---|
| Output resolution tiers | 1K, 2K, 4K | 1K, 2K | 1K, 2K, 4K | 1K, 2K, 4K |
| Primary strength | Multi-reference scenes, native 4K | Typography, precise edits | Fast generation, everyday edits | Precision edits preserving approved parts |
| Choose it when | Multi-reference scenes, 4K finish | Posters and typography-heavy edits | Fast exploration, high volume | High-stakes single edits |
| Total input images | Up to 10 | Up to 51 source + 4 guide references | Up to 16 | Up to 16 |
| Images per request | Up to 1 | Up to 4 | Up to 4 | Up to 4 |
| Output formats | PNG, JPEG | Provider default | PNG, JPEG, WEBP | PNG, JPEG, WEBP |
| Selectable quality | No quality selector | low, medium, high | low, medium, high, xhigh, max | low, medium, high, xhigh, max |
| 1K generation / editing | 10 / 10 creditsQuality: provider controlled | 12 / 12 creditsQuality: medium | 2 / 3 creditsQuality: low | 2 / 3 creditsQuality: low |
| Extra reference cost | Included in the quote | Included in the quote | 1 included; +2 cr per extra image | 1 included; +2 cr per extra image |
| Explore the model | FLUX 3 Image | Ideogram 4.5 | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst |
Generate from text or upload references to edit. Your selected resolution determines the credit quote; the reference-input limit is separate from output resolution.
FLUX 3 is Black Forest Labs' multimodal model family. FLUX 3 Image handles still-image generation and editing. Earlier FLUX 3 releases and guides also cover video; creating a clip requires a separate video workflow.
Yes. The workspace on this page starts with FLUX 3 Image. Write a prompt to generate a new image, or upload references and describe an edit. The displayed examples are official BFL examples, not Oeneo AI test results.
You can provide up to 10 images in one FLUX 3 Image editing request. Describe which image supplies each subject, product or setting. With Auto aspect ratio, the first reference determines the output shape, so keep upload order consistent with your instructions.
You can download a result and upload it for the next edit. State only the new change and inspect the parts you want to preserve after each round. This is not an automatic editing session; the fal reference-size limit also applies to every re-upload.
BFL's official Playground demonstrates bounding-box composition and editing. This page does not include that canvas or a mask editor. You can describe positions and relationships in a prompt, but that is different from direct visual layout control.
Not through the current fal endpoint if the image exceeds 4 MP. Output resolution and reference-input limits are separate. Resizing a large image also changes its pixels, so original-resolution, pixel-identical editing should not be assumed.
BFL describes multilingual text generation, and its official examples include lettering inside images. Chinese text quality has not been verified in Oeneo AI. Keep copy short and proofread every character, name and date.
FLUX 3 Image uses credits based on the resolution you select. The current rates are shown in the specifications and the workspace quote. Registration credits are limited, and free-user generation results are public. Paid accounts can keep results private.
In the October 1, 2026 launch announcement, BFL described open weights as a future release and commercial weights as available by arrangement. Check BFL for the latest licensing status. Hosted image creation and licensing weights for deployment are separate decisions.
Use the existing Oeneo AI tools for the next step after generation.
Use the FLUX 3 image generator for a new composition, or upload references for a focused FLUX 3 image edit in Oeneo AI.
