Update WaveSpeed inference provider description

[chengzeyi]

Hi! We're the WaveSpeed team (an Inference Provider on the Hub). This PR updates our provider description on docs/inference-providers/providers/wavespeed.md, edited via the wavespeed.handlebars template as the file header instructs, with the generated .md updated to match (same approach as #2027).

Why

The current blurb is generic and doesn't tell a reader anything about what we actually serve:

WaveSpeedAI is a high-performance AI inference platform specializing in image and video generation. Built with cutting-edge infrastructure and optimization techniques, WaveSpeedAI provides fast, scalable, and cost-effective model serving for creative AI applications.

Proposed

WaveSpeedAI is an inference platform for image and video generation. On Hugging Face it serves open-weight model families including FLUX, Qwen-Image, Z-Image, HiDream, Wan, HunyuanVideo, LTX-2 and MiniMax-H3, covering text-to-image, image-to-image, text-to-video and image-to-video from a single API, as well as LoRA adapters hosted on the Hub. The wider WaveSpeedAI catalog also includes closed-weight models such as Seedream, Seedance, Nano Banana and GPT Image.

We aimed to match the tone and length of the other provider templates (fal-ai, replicate, together, baseten) — factual, no superlatives, no performance claims.

Evidence for the claims

From https://huggingface.co/api/models?inference_provider=wavespeed&expand[]=inferenceProviderMapping, counting live mappings where provider == "wavespeed":

Task	single-model mappings	tag-filter (LoRA) mappings
text-to-image	13	44,645
image-to-image	8	451
text-to-video	5	155
image-to-video	6	176
image-text-to-video	1	12
image-text-to-image	0	10

The 33 single-model repos behind the named families: FLUX.1-dev / -schnell / -Krea-dev / -Kontext-dev, FLUX.2-dev / -klein-*, Qwen-Image / -2512 / -Edit / -Edit-2509 / -Edit-2511, Tongyi-MAI/Z-Image (+ Turbo), HiDream-I1-Dev / -Full, Wan2.1/2.2 T2V, I2V and TI2V, tencent/HunyuanVideo / -1.5, tencent/HunyuanImage-2.1, Lightricks/LTX-2, MiniMaxAI/MiniMax-H3, ByteDance/lynx, chetwinlow1/Ovi, briaai/FIBO, ideogram-ai/ideogram-4-fp8, krea/Krea-2-Turbo.

Scope

Deliberately minimal — copy only, two files, no other changes:

• Partners table in index.md: untouched. Its columns are Chat completion, Feature Extraction, Text to Image, Text to video, Speech to text — there's no image-to-image or image-to-video column, and other image/video providers are equally uncredited there. Mentioning it only as an observation; adding columns is a product decision for you, not something a vendor PR should push.
• Featured code snippets: untouched. Our text-to-video snippet currently points at larryvrh/MiniMax-H3-Turbo-Lora, a third-party user repo rather than MiniMaxAI/MiniMax-H3. We checked generate.ts and this is fully automated (sort=likes30d, top warm model per task/provider), and that repo genuinely leads by likes30d (750 vs 328 for the runner-up) — so it's working as designed and not ours to override. Flagging it only in case you consider it surprising that a community LoRA fine-tune is the featured example for a provider.
• Prettier: .handlebars is in .prettierignore, and docs/ is outside the scripts/inference-providers format scope. prettier --check on wavespeed.md reports the same pre-existing warning before and after this change.

Happy to adjust the wording however you'd prefer. Note: this PR was prepared with AI assistance (Claude Code); all figures above were verified against the Hub API and the numbers/claims were reviewed by us.

---

Note

Low Risk
Documentation-only wording change for a provider page; no code, API, or security impact.

Overview
Replaces the generic WaveSpeedAI blurb with a factual description of what the provider serves on Hugging Face: named open-weight image/video families, Hub LoRAs, and a note that the wider WaveSpeed catalog also includes closed-weight models.

The copy is updated in wavespeed.handlebars and the generated wavespeed.md. No task snippets, logos, or partner-table changes.

^{Reviewed by Cursor Bugbot for commit 8eeeb35. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8eeeb35fcd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Remove the unsupported image-to-video API claim

The current integration generator does not include image-to-video among its supported TASKS (scripts/inference-providers/scripts/generate.ts:8-28), and this generated provider page consequently exposes only image-to-image, text-to-image, and text-to-video. Users of the documented Hugging Face Inference Providers API therefore have no image-to-video task/helper through which to use the capability promised here; either add first-class task support before making this claim or describe image-to-video as part of WaveSpeed's wider catalog instead.

Useful? React with 👍 / 👎.