Black Forest Labs’ FLUX led image-generation activity on Poe in the platform’s early-2025 data, taking close to 40% of image messages. Poe also reported that the combined relative share of DALL-E 3 and various Stable Diffusion versions fell by nearly 80% as its official image-model catalog expanded from about three models to approximately 25.
That is a substantial shift on Poe, not proof that DALL-E lost 80% of the global market, users, revenue or image-generation volume. Poe’s report, published March 10, 2025, measures normalized message shares among Poe subscribers over roughly the preceding year. (Poe’s report)
What Poe actually measured
Poe operates as a multi-model platform. Its figures count image-generation messages sent through Poe, rather than activity across every consumer app, API, enterprise deployment or local installation.
- Metric: messages or interactions, not revenue, subscriptions, unique users, image files, compute consumption or profit.
- Population: Poe subscribers, not all users of ChatGPT, Midjourney, Adobe Firefly, Google’s products, direct FLUX APIs or locally run Stable Diffusion.
- Method: weekly percentage shares normalized to account for subscriber growth.
- Catalog change: Poe says its official image-model selection grew from roughly three to approximately 25 models.
That last point matters. When many new choices enter the same platform, older models can represent a much smaller percentage of activity even if their absolute message count is flat or rising. Poe’s chart therefore describes a change in the composition of Poe usage, not a complete industry census.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →#1 Best Overall
The report says Poe had more than 100 text, image, video and audio models overall. Its image figures should be read as a platform-specific snapshot, with the date and denominator attached.
The figures behind the headline
| Model family or category | Poe-reported early-2025 result |
|---|---|
| FLUX | Close to 40% of image-generation messages |
| Imagen 3 family, including Imagen 3 and Imagen 3 Fast | Almost 30% |
| Playground and Ideogram combined | Approximately 10% |
| DALL-E 3 plus various Stable Diffusion versions | Combined relative share down nearly 80% from earlier levels |
The final row is a decline measurement, not a current percentage share. The categories are also not directly equivalent: FLUX and Imagen are reported as current shares, while the DALL-E/Stable Diffusion figure describes how a grouped share changed over time. All figures are approximate and apply to Poe’s image traffic.
Did DALL-E itself fall 80%?
No conclusion that specific is supported. Poe grouped DALL-E 3 with multiple Stable Diffusion versions when it described a nearly 80% fall in relative share. The report does not show that DALL-E alone:
- lost 80% of its absolute messages;
- lost 80% of global users;
- generated 80% less revenue; or
- was displaced in every distribution channel.
Four different concepts are often collapsed into one headline:
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitches- Relative share: a model’s percentage of messages within a defined platform and period.
- Absolute volume: the number of messages sent.
- Revenue share: money paid to a provider.
- Global market share: activity across consumer products, APIs, enterprises and self-hosted systems.
Only the first concept is established by Poe’s report. For example, if an older model handled 70 of 100 image messages and later handled 70 of 500, its volume would be unchanged while its share would fall from 70% to 14%. That arithmetic example is illustrative, not Poe’s underlying count.
What “FLUX dominates” means here
Within Poe’s image category, FLUX emerged as the leading family, with close to 40% of messages. Calling that “dominance” without naming Poe’s platform overstates the evidence. The data does not establish that Black Forest Labs has 40% of worldwide AI-image usage or that FLUX is the world’s most-used image model.
FLUX entered the market in mid-2024. Its rapid adoption may reflect a combination of prompt adherence, visual quality, distribution through hosted services, multiple model variants and an open-weights route that supports experimentation outside a single consumer application. Those are plausible strategic factors, not causes proven by Poe’s chart. Third-party platforms making FLUX easy to access could also have amplified switching.
Imagen makes this more than a FLUX-versus-DALL-E story
Google’s Imagen 3 family reached almost 30% of Poe image messages, making it a major challenger in the same dataset. The more accurate picture is a three-part shift: FLUX led, Imagen built a large second position, and early models collectively lost relative share as the menu expanded. Playground and Ideogram together retained roughly 10%, while the remainder went to DALL-E, Stable Diffusion variants and smaller offerings.
This is why a headline focused only on DALL-E misses the market structure. Aggregators reveal users switching among several model families, not simply abandoning one vendor for another.
What the shift says about AI markets
Distribution can matter as much as benchmarks
A model available through an existing workflow can gain usage quickly. Poe’s data records the result of that availability, but it does not isolate whether quality, speed, text rendering, licensing or price was decisive.
New models dilute incumbent share
Catalog expansion changes the denominator. A falling percentage is a warning about competitive pressure, not automatic evidence of a matching fall in demand.
Leadership differs by channel
Poe subscribers, direct API customers, enterprise buyers and local-model users have different incentives. A ranking in one channel should not be exported to all others.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Dates and denominators are essential
Poe published the report on March 10, 2025, using activity from approximately the prior year. It is not August 2026 market-share data, and it cannot establish a current global ranking without newer, broader evidence.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to choose a model for real spending decisions
For a business or creator, the Poe ranking is a starting signal rather than a purchasing decision. Evaluate:
- Prompt adherence: complex spatial, stylistic and compositional instructions.
- Text and anatomy: posters, packaging, logos, people, hands and product photography.
- Editing and consistency: preservation of characters, products, layouts and reference images.
- Resolution and aspect ratios: suitability for print, social platforms, commerce and video.
- Latency and throughput: production speed under your expected workload.
- Rights and privacy: commercial permissions, training-use policies, indemnity, data retention and reference-image handling.
- Tooling: SDKs, rate limits, webhooks, batch jobs, editing endpoints and deployment choices.
- Cost predictability: subscription credits, per-image billing, API tokens and infrastructure costs.
| Need | Candidate | Why it may fit | Main financial or operational caution |
|---|---|---|---|
| Compare many models | Poe | Several model families in one interface | Its internal usage shares are not the whole market; advanced plans start at $4.99 per month on Poe’s current About page, while availability and pricing can change. (Poe) |
| Direct FLUX production access | Black Forest Labs | Pay-as-you-go API plus enterprise options | Model-specific licenses, hosting costs and support terms require review. The company advertises no subscription or seat fees, volume discounts, SLAs and dedicated support. (Black Forest Labs pricing) |
| Existing OpenAI stack | OpenAI image API | Fits broader OpenAI infrastructure | The current pricing documentation lists GPT-image-2 rather than DALL-E 3; exact image cost depends on token usage and request settings. (OpenAI API pricing) |
| Bundled consumer creative tools | Google Flow | Free access and paid Google AI plans | Displayed prices, credits and availability vary by market; the page shows Plus at $4.99 per month and Pro at $19.99 per month. (Google Flow) |
| Local or self-hosted deployment | FLUX open-weight options | More control over infrastructure and workflows | GPU, maintenance, compliance and licensing obligations move to the buyer; open weights do not automatically mean unrestricted commercial use. |
What buyers should take from the numbers
FLUX’s Poe result is commercially relevant because it signals strong adoption in a multi-model environment. It does not eliminate the need to test output quality, consistency, latency, rights and total cost for a particular workflow. OpenAI’s current image lineup is also broader than the legacy DALL-E 3 label, while Google’s bundled plans may suit experimentation but make per-image costs harder to compare.
For production, calculate the full cost: generation fees, retries, editing, storage, moderation, integration work, GPU or cloud infrastructure, and the labor required to review outputs. A model with a smaller platform share can still be the cheaper or safer choice for a specialized use case.
Free tools Windows power users keep installed
One-click scans. No signup required.
The precise conclusion
Poe’s March 10, 2025 report found FLUX leading image-generation messages on Poe at close to 40%, with Imagen 3 near 30%. It also found the combined relative share of DALL-E 3 and older Stable Diffusion versions down nearly 80% as Poe’s catalog expanded. That is evidence of rapid platform-level substitution and market fragmentation—not proof of an 80% global DALL-E collapse or 40% worldwide FLUX ownership.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
