FLUX AI image generator is a model-family description, not one universal app. Black Forest Labs offers a browser Playground as well as an API, and other services may host FLUX with their own controls and billing. For a first image task, identify the provider and the exact model before comparing results or buying credits.
If the goal is an original character you can also interact with, Cherrypop’s character creator combines appearance with personality and conversation setup. FLUX is the more relevant topic when you are choosing an image-generation or reference-editing workflow.
Start with the actual FLUX surface
BFL’s documentation introduction directs users to either the Playground or an API quick start. It keeps FLUX.2 and FLUX.1 Kontext documentation separate from the newer FLUX 3 family. An article about one version should not be treated as a specification for every option carrying the FLUX name.
| Access route | Useful for | Check before relying on it |
|---|---|---|
| BFL Playground | Trying a documented model in a browser | Model, available input controls and displayed charge |
| BFL API | Integrating generation into software | Exact endpoint, credentials, billing and result handling |
| Another provider hosting FLUX | Using FLUX inside that provider’s workflow | Its model version, reference limits, moderation and export terms |
This guide follows BFL’s browser documentation. It does not establish another host’s free allowance or promise that a downloaded model has the same controls as the hosted service.
Create one image before designing a whole collection
The Playground guide describes selecting a model, entering a prompt and generating. For editing, it describes uploading a source image and stating the change. It also documents downloading the image and copying API Code from a successful generation. That code path is optional; it is not required for a browser experiment.
For an original test brief, imagine a fictional travel poster showing a red cabin beside a lake. Decide whether the job is a fresh interpretation or a controlled revision of a cabin image you already own. Those goals should not share the same acceptance rule.
| Task | Success means | A misleading success measure |
|---|---|---|
| Explore a new concept | A usable composition and visual treatment | The output resembles an unrelated polished gallery example |
| Move an existing cabin into a new setting | Identifying cabin details remain suitable | The new landscape looks attractive but the cabin changes |
| Change the poster palette | The important shapes remain while colors change | Every element is regenerated into a different design |
Give each reference a single job
BFL’s FLUX.2 editing guide documents multi-reference input, with different limits for API and Playground. Its model overview further distinguishes reference capacity and controls by variant. A maximum reference count is a capacity limit, not a recommendation to fill every slot.
For the cabin poster, use the following original reference plan. Begin with the fewest sources that communicate the required decisions, and remove contradictory inputs before generating.
| Reference | Assigned role | What it should not decide |
|---|---|---|
| Your cabin photograph | Shape and identifying architectural details | The final poster’s complete composition |
| A lake composition sketch | Where land, water and open space belong | A replacement design for the cabin |
| Your own color sample | The intended palette | New objects or extra buildings |
Describe those roles in the request. Then inspect the output against each role separately. If the cabin gains an extra window, that matters more than whether the surrounding forest looks impressive. If a source is only inspiration, allow interpretation; if it represents a real item, be stricter about fidelity.
Choose controls for a reason
The FLUX.2 overview distinguishes standard controls from the adjustable steps and guidance offered by flex, and identifies grounding search as a max feature. Those controls serve different tasks; choosing a variant still depends on what your image needs to do.
Start with one model and a small repeatable brief. Change the model only when you have a reason to evaluate a different control or cost. If you change the prompt, reference set and model together, you cannot tell which decision improved the result.
Grounding does not remove the need for verification. An image containing a date, label, map or product feature can be persuasive and still wrong. Confirm those details outside the image before using them as information.
Keep spending and accepted work visible
BFL’s Playground guide says Playground and API pricing are the same. That does not make every model or reference configuration the same price. Check the selected operation before submission, and keep completed outputs you reject in your project cost record.
A simple record needs the provider, model, inputs, requested change, displayed charge and acceptance reason. Save the approved file separately from exploratory versions. If you later move to an API, the successful browser settings are a starting specification, not proof that your integration is tested.
When to switch from an image to a companion character
A portrait or scene can remain a standalone asset. When you want personality and interaction to accompany the visual, Cherrypop’s creator offers realistic and anime styles, character details, a scenario and an opening message. You can continue into chat and supported media creation rather than treating the character as only an image file.
That recommendation is about task fit. It is not a claim that Cherrypop exposes BFL’s API, uses a particular FLUX version, or guarantees the same character appearance in every generation.
Cherrypop is free to start with limits; relevant features may require Premium or Cherries. Create an original character if interaction is your next goal, and check the displayed media cost before generating.