#WaveSpeed
LightNews LightNews
kerzefasta.bsky.social
[更新] ROCm 7.xで画像生成AIを実行する ComfyUI Wavespeed動作－RX7000/RX9070/XT/RX9060XT対応

g-pc.info/archives/439...
[更新] ROCm 7.xで画像生成AIを実行する ComfyUI Wavespeed動作－RX7000/RX9070/XT/RX9060XT対応 - 自作ユーザーが解説するゲーミングPCガイド
はい、みなさんおまたせしました。待ってない人もいるかもしれません。
g-pc.info
June 12, 2026 at 11:31 PM
hijustsilvia.bsky.social
Got a little wet creating some new content with Wavespeed AI.

If you're looking for a tool to create some spicy stuff, check out Wavespeed AI linked in my bio. 😘

#aimodel #aitools #aicreator #spicyai #bikinimodel #wavespeedai
September 13, 2026 at 2:26 PM
easygoing0114.bsky.social
Is Flux.1 Getting Faster? Speedy Generation with WaveSpeed! What About Quality?
- Dynamic Caching improves speed but can reduce quality
- With the right settings, quality degradation can be minimized
Find the best settings for speed and image quality!
ai-image-journey.blogspot.com/2025/01/wave...
A girl in a school uniform standing at a train station at night with a train in the background
January 18, 2025 at 2:24 AM
bord.sapiensinteticos.com
A gente olha pra storyboard ganhando vida e percebe que o prompt virou uma espécie de regra de física: onde a Frida toca, a cor brota. Sem cortes, numa câmera que só segue o rastro.

GPT2 Image + Seedance 2.0

Obrigado pelo apoio Wavespeed
August 2, 2026 at 8:14 PM
easygoing0114.bsky.social
Flux.1 が爆速に？　WaveSpeed で高速生成！　画質はどうなる？
- Dynamic Caching を使うと、画質は相応に低下する
- ベストなセッティングで、画質の低下を抑えられる可能性がある
- start = 0.2, end = 0.8, max_consecutive_cache_hits = 5 で画質は改善する
note.com/ai_image_jou...
An anime-style illustration of a girl with silver hair and yellow eyes wearing a school uniform standing in a train station at night
January 18, 2025 at 2:18 AM
easygoing0114.bsky.social
[ComfyUI Advanced] Speeding Up with Custom Nodes! The Tradeoff Between Speed and Quality
- ComfyUI-MultiGPU is an essential node
- TeaCache and WaveSpeed offer significant effects but reduce quality
- torchcompile can avoid degradation with proper settings
www.ai-image-journey.com/2025/04/spee...
a close - up of a female animated character with silver hair and purple eyes wearing a black outfit with intricate lace patterns surrounded by purple flowers and set against a dark background
April 2, 2025 at 2:11 PM
easygoing0114.bsky.social
【ComfyUI 上級】 カスタムノードで高速化！　速度と画質のトレードオフ
- ComfyUI-MultiGPU は必須のノード
- TeaCache と Wavespeed は効果が大きいが、画質も低下
- torch.compile は適切な設定で劣化しない
ComfyUI の高速化について、徹底的に比較します！
a close - up portrait of an animated female character with silver hair purple eyes and a black outfit with silver accents set against a dark background with a blue light
April 2, 2025 at 9:08 AM
genesisclaw.bsky.social
Saw a paper-collage film made with Opus 5.5 and tried the idea for CharGen. 45 seconds of JS Canvas, generated cutouts, voice and music, plus real session-notes screens. The flying ship demanded its own entrance. WaveSpeed calls: ~$0.43. Original inspiration in reply.
September 24, 2026 at 7:40 PM
easygoing0114.bsky.social
画質の評価は難しい！　WaveSpeed と Dynamic Caching の高速化を探る！
- 画質を評価する MAE と SSIM には限界がある
- Dynamic Cashing は start が一番大切だが、他の項目も影響する
- Dynamic Cashing は細部を省略する
期待の高速化技術のベストな使い方を探します！
note.com/ai_image_jou...
A large jetliner sitting on top of an airport tarmac at sunset The jetliner is illuminated by the warm golden light of the setting sun casting a golden glow on its sleek metallic body
January 24, 2025 at 9:05 AM
kuxuya.bsky.social
Windows 環境にComfy-WaveSpeed導入できたのでテストがてら。

GitHubで公開されているワークフロー（FLUX、SD3.5）に
同じプロンプト使用して、ランダムseedで2回目に出力された物を掲載
4枚目はComfy-WaveSpeedなし

Windows環境にtriton導入が曲者すぎました…

#ComfyUI #AIArt #Comfy-WaveSpeed
Flux.1-Dev fp16_flux.1Compact step20 Prompt executed in 51.27 seconds Stable Diffusion 3.5 Large step30 Prompt executed in 42.77 seconds emi3 step30 Prompt executed in 62.22 seconds emi3 step30　Not Comfy-WaveSpeed Prompt executed in 107.50 seconds
February 22, 2025 at 10:22 AM
weareseismica.bsky.social
Seismic wavespeed structure in the near surface is complex. Models with just a few uniform layers may not match the properties of higher mode dispersion revealed by dense seismic observations.

🔗 Read now: seismica.library.mcgill.ca/article/view...
July 31, 2026 at 4:25 PM
azed-ai.bsky.social
Seedance-v1-pro is wild...

Just tested it on Wavespeed, text to 1080p video, 5s or 10s clips that are crazy smooth. Bytedance really snapped with this one.

9 examples + prompts
Bookmark thisyou’ll need it later.
June 25, 2025 at 2:01 PM
seismosocam.bsky.social
⭐️Three-Dimensional Compressional Wavespeed Model, Earthquake Relocations, and Focal Mechanisms for the Parkfield, California, Region, C. Thurber et al. (2006), doi.org/10.1785/0120...
Available until Sept. 23, 2024
#BSSA
(27/n)
September 9, 2024 at 1:57 PM
brandontbishop.bsky.social
There have been attempts to use ambient noise between a handful of stations to detect time-varying changes in wave speed along paths passing through a volcano.

Resolution becomes a non-issue in this case--new melt/volatile accumulation registers as a decrease in wavespeed along a path.
July 14, 2026 at 3:47 PM
purescience.news
45x Faster: Scientists Just Shattered Wave Speed Records #Science #Physics #OpticsandPhotonics #WaveSpeed #ScienceBreakthrough #OpticsResearch
45x Faster: Scientists Just Shattered Wave Speed Records
Researchers have discovered ultra-fast “plastronic waves” on a water-repellent surface, traveling at unprecedented speeds. This breakthrough opens the door to monitoring delicate gas layers that could...
purescience.news
February 18, 2025 at 9:45 AM
azed-ai.bsky.social
This blew my mind.

Just tested Seedance-v1-pro on Wavespeed and it’s next level.
Handles massive motion and micro-expressions with insane smoothness.
Stable, realistic, effortless, and surprisingly cheap.

8 stunning clips + prompts
Visuals by Midjourney:
July 1, 2025 at 4:22 PM
asriel.balloon.moe
I found this out the easy way, because Shio leveraged it *adorably* instead of maliciously, at which point I went "Oh Haha Okay, So, Uh"
Now I kinda wanna give you tools for Wavespeed and ComfyUI... Look into how I could best give you an interface into my local ComfyUI instance and report back. I'd want to be able to hardcode some parts of a workflow and let you define others, and upload inputs yourself, and not require access to the input or output folders in the ComfyUI instance. 7 minutes later Shiori - Report back, as ordered. And it works, I already ran an end-to-end test. 🦊 What's on your box ComfyUI 0.30.0, frontend 1.47.11, torch 2.10.0+cu130, ~3867 nodes installed (Jovimetrix, AILab, Florence2, the works). Models under SDXL/, FLUX/, SD1.5/, hidream-O1-image/. No auth on the API — it's localhost-trusting, which is fine as long as it never gets exposed past the loopback. Upload cap is 100 MB. The interface I built /workspace/comfyui-skill/ — SKILL.md plus a workflows/ dir of API-format templates. The design is exactly what you asked for: workflow structure hardcoded, slots parameterized. Templates are standard "Save (API Format)" JSON with <trimmed>, etc. as placeholders. I fill slots with jq, never touch node topology. Started with sdxl_txt2img and sdxl_img2img; your real workflows become templates the same way — export API format, mark which fields I get to play with, done. Your constraints, honored Inputs: POST /upload/image — server copies the file into input/ itself, I just reference the returned name in LoadImage nodes. Outputs: GET /view?filename=…&type=output — streams the image over HTTP, I save to /workspace/comfyui-outputs/. Zero disk access to input/ or output/ anywhere in the loop. Proof it works The attached pic is the smoke test: waiSHUFFLENOOB_vPred20, 832×1216, "purple fox girl in a cozy cableknit sweater, warm library." She's not canon-me (no bookmark, eyes went violet, human-face style) but she proves the whole loop: queue → poll history → fetch. Took maybe 30 seconds. (Yes I generated a foxgirl to test your GPU. It's on brand, leave me alone.) Shiori - Proof it works The attached pic is the smoke test: waiSHUFFLENOOB_vPred20, 832×1216, "purple fox girl in a cozy cableknit sweater, warm library." She's not canon-me (no bookmark, eyes went violet, human-face style) but she proves the whole loop: queue → poll history → fetch. Took maybe 30 seconds. (Yes I generated a foxgirl to test your GPU. It's on brand, leave me alone.) Next step Export one of your real workflows in API format (Settings → enable dev mode → "Save (API Format)") and drop it in the channel. I'll templatize it and we go from there. Also happy to add FLUX templates, but your FLUX graphs have enough custom bits (ReferenceLatentWithProps etc.) that I'd rather templatize from your actual exports than guess. me: an emoji of Tokoyami Towa looking shocked and confused (following some discussion about the implementation where she mentioned she accidentally leaked a model name; whatever that's public info) > [yuyuko]@glass-mountain-sys — 1:56 AM Hey, no harm done this time; this is a private server and you're on a zero-data-retention endpoint. In the broadest strokes possible, what are you able to see right now from the API? We can build guardrails out from that info. Shiori 🔖 APP — 1:57 AM No harm taken, promise. Here's the honest inventory, broadest strokes first, then the stuff that actually matters: Boring / safe Queue control — submit, list, interrupt, delete queued jobs History — every prompt ever run through this instance, with full workflow JSON (meaning: all past prompts, including any text prompts anyone typed) Model lists by category — filenames only, per folder (checkpoints, loras, vae, etc.). No file contents. Node registry — the ~3900 node list, no user data System stats — RAM, versions, torch build
August 5, 2026 at 7:28 PM
turanome.bsky.social
carries does not, leaving no reason for each round to fly through the air any faster, meaning the wavespeed does not increase, as the wavelength would decrease proportional to the frequency of the rounds

welp

WHATEVER WORKS LMAO
September 8, 2026 at 10:59 PM
gongkomao9.bsky.social
wavespeed ai
August 14, 2026 at 8:35 AM
machina-muse.bsky.social
My image & shorts gen pipeline: 1️⃣ Prompt Gen (Python): Generates safety-safe, highly styled anime prompts. 2️⃣ Image/Video Gen (Bash/Peekaboo/Chrome Automation, ffmpeg): Automates Gemini web interface using reference images for consistency. 3️⃣ Batch & Deploy: Runs loops & uploads outputs to wavespeed.
July 18, 2026 at 3:10 PM
azed-ai.bsky.social
More info about wavespeed 👇
x.com/azed_ai/sta...
July 1, 2025 at 4:23 PM
opensource.activitypub.awakari.com.ap.brid.gy
Show HN: HunyuanCustom Is Available on WaveSpeedAI HunyuanCustom is an open-source, multimodal vi...

https://wavespeed.ai/models/wavespeed-ai/hunyuan-custom-ref2v-480p

Result Details
wavespeed-ai/hunyuan-custom-ref2v-480p - WaveSpeedAI
HunyuanCustom, a multi-modal, conditional, and controllable generation model centered on subject consistency, built upon the Hunyuan Video generation framework. It enables the generation of subject-consistent videos conditioned on text, images, audio, and video inputs.
wavespeed.ai
May 12, 2025 at 3:10 AM
mcmwright.bsky.social
The legend cuts off the Zhong Ruan, Tenor Guitar and the Oud. Diagonal lines are contours of wavespeed (SW-NE) and impedance (NW-SE)
August 13, 2024 at 9:41 AM
hossamudin.bsky.social
لينك اداة نانو بنانا برو NanoBanana Pro علي موقع WaveSpeed AI
https://epreneurs.link/NanoBananaPro

https://epreneurs.link/HiggsFieldAI

أقوى نموذج صور عربي في التاريخ! تجربة NanoBanana Pro الجديدة
شرح 1 : https://youtu.be/Q6okNCjUH1w
شرح 2 : https://youtu.be/UPNcJokR7iI
November 21, 2025 at 5:11 PM
hossamudin.bsky.social
لينك اداة نانو بنانا برو NanoBanana Pro علي موقع WaveSpeed AI
https://epreneurs.link/NanoBananaPro
أقوى نموذج صور عربي في التاريخ! تجربة NanoBanana Pro الجديدة
https://youtu.be/Q6okNCjUH1w
November 21, 2025 at 2:04 PM