FLUX 3 视频发布:性能宣传尚不足以证明全面领先
现有资料显示 FLUX 3 Video 已面向 API 和部分合作伙伴开放,但“比 Seedance 2.5 与 MiniMax H3 加起来还强”的说法缺乏独立证据支持。
Black Forest Labs 的 FLUX 3 Video 已公开可用,支持文本、图像和视频驱动的视频生成,并包含原生音频能力。公开报道转述了该公司自有测试中的排名表现,但比较对象包括 Seedance 2.0,而非候选消息所称的 Seedance 2.5。
现有对比更接近于“工作流取舍”而不是单一模型全面胜出。资料称,FLUX 3 在文字渲染、风格化画面、关键帧控制和单位时长成本方面可能更有优势;Seedance 更适合较长连续镜头及复杂参考素材控制;MiniMax H3 则被描述为侧重多模态输入和开放权重生态。
因此,社交媒体中的绝对化结论应视为主观宣传。不同来源还混用了 Seedance 2.0 与 2.5,且部分价格与规格来自第三方平台,发布前应以各厂商的官方文档、版本号和独立测试为准。
来源证据
Flux 3 vs Seedance 2.5: Duration, Cost, Controls Comparedaireiter.com · supportingIf your production hinges on longer uninterrupted clips, 4K resolution, or reference-video control, Seedance 2.5 is the stronger pick despite the higher cost. If you need typography, stylized aesthetics, precise keyframe control, or the lowest per-second rate, Flux 3 wins-and its agentic chaining can close the duration gap for multi-shot sequences. The trade-off that remains unresolved is whether Flux 3's chaining seams will hold up at scale as cleanly as Seedance 2.5's in-pipeline extensions-a question that only sustained production testing will answer. Related reading: Flux 3 overview Seedance 2.5 overview Seedance 2.5 vs Kling 3.0 video comparison Flux 3 vs Minimax H3 video comparison [...] Flux 3's reference system emphasizes explicit control points-keyframes, video continuation, and video-to-video-while Seedance 2.5 centers on reference-video control and post-generation audio-visual editing. The split matters when your workflow depends on precise keyframe transitions (Flux 3
Black Forest Labs makes FLUX 3 Video generally available and claims it beats Seedance 2.0the-decoder.com · supportingBFL's own tests put FLUX 3 at the top of the Elo rankings for text-to-video with a score of 1,135 and image-to-video with 1,051. It ranks ahead of Gemini Omni Flash, Minimax H3, and Seedance 2.0. Pricing is based on each second of video output. Draft mode is limited to HD and costs $0.06 per second for text-to-video or image-to-video, and $0.12 for video-to-video. At full quality, HD costs $0.17 per second for text-to-video or image-to-video, and $0.41 for video-to-video. Full HD costs $0.29 and $0.53 per second, respectively. Audio is included. More video examples are available on the BFL blog. Ad DEC\_D\_Incontent-1 Ad ### AI News Without the Hype – Curated by Humans [...] Aug 5, 2026 AI company Black Forest Labs (BFL) has made its FLUX 3 Video model generally available through the BFL API and select partners. The model generates HD and Full HD clips up to 20 seconds long, with native audio that includes dialogue, sound effects, and ambient noise. It supports text-to-video, ima
Seedance 2.5 vs MiniMax H3: Which Video Model Wins? - GlobalGPTglbgpt.com · supporting### 3. Seedance accepts a much larger reference set Seedance 2.5 supports up to 30 images, 10 videos, and 10 audio clips in a single request. MiniMax H3 supports up to 9 images, 3 videos, and 3 audio clips, capped at 12 files across input types. These limits do not prove that one model understands every reference better, but they do change the kind of project each workflow can comfortably organize. [...] MiniMax H324 FPS Audio Seedance 2.5Joint audio-video generation MiniMax H332 kHz native stereo audio Reference inputs Seedance 2.5Up to 30 images, 10 videos, and 10 audio clips MiniMax H3Up to 9 images, 3 videos, and 3 audio clips; 12 files maximum across input types Editing controls Seedance 2.5Timestamp control, targeted audio/video edits, green-screen editing, viewpoint and camera edits, and reference editing MiniMax H3Generalized multimodal reference and editing; task support also depends on the selected checkpoint or hosted workflow Open weights Seedance 2.5No open-wei
Stephen Day | AI Agents + Vibe Coding | Codex (@passivelywealthydad) on Threadsthreads.com · supportingTranslate Image 3 2 Image 4: ninetiespope's profile picture ninetiespope 2d Minimax H3 started a new era in open source video diffusion. WAN and LTX family gave us countless hours of fun, but it's time to turn the page and accept the new SOTA open source model. What a time to be alive. Will see what Flux 3 brings to the table, but for now, H3 is the new open source king. The workflows and things community is gonna make will be amazing. #MiniMaxH3 #hailuoh3 #videodiffusion Image 5: vincent.chanw's profile picture vincent.chanw 4d MiniMax H3 和 Seedance 2.5 走的是兩條路。 H3 主打 omni-modal + open weights:文字、圖片、影片、音訊可放進同一個 context,生成最高 15 秒、2K、native stereo audio;價格亦壓到主流 2K 模型的三分之一以下。 Seedance 2.5 的優勢則是導演能力:30 秒長序列、角色一致性、因果劇情、camera movement 與 shot plan 執行,現階段成片效果明顯更強。 [...] []( []( []( # Thread 420 views Image 1: passivelywealthydad's profile picture passivelywealthydad AI Threads 2d Flux 3, MiniMax H3 and Seedance 2.5 have arrived and the future of AI video is alive and well.
류내원 (@won.wizard) on Threadsthreads.com · supporting[]( []( []( Log in # Thread 330 views Image 1: won.wizard's profile picture won.wizard 20h Seedance 2.0 vs MiniMax H3 vs FLUX 3 Video Price Comparisons 최신 발표된 MiniMax H3, FLUX 3 와 Seedance 2.0 가격 비교해 보았습니다. 가격은 Fal ai 가격 기준입니다. Seedance 2.0 : 720p $0.30/sec, 1080p $0.68/sec. Flux 3 : 720p $0.17/sec, 1080p $0.29/sec MiniMax H3: 768p $0.16/sec, 2K $0.26/sec based: Fal ai #Seedance2 #MiniMaxH3 #FLUX3 #aivideo #wonwizard Translate Image 2 2 Related threads Image 3: passivelywealthydad's profile picture passivelywealthydad AI Threads 2d Flux 3, MiniMax H3 and Seedance 2.5 have arrived and the future of AI video is alive and well. H3 is blowing me away right now and is currently uncensored generating a ton of copyright material. Better jump on it before it’s gone! 1
Building an AI Music Video with FLUX 3, Seedance 2.5 + MiniMax H3youtube.com · supportingcool. It's a very long video, so I don't I don't I don't want to waste your time. It's a very long video. No, no, no. This is uh Is this posted anywhere, though? No, no, no, no. It's not posted anywhere yet. What is Tell me about this for a pitch for for pitching. Oh, you're pitching. It's a supernatural thriller. Oh, okay. Okay. Very nice. It looks cool. It's looks like it's really well done. It's uh getting the realism down pretty good. Looks very believable. Is this Is this in sea dance? Uh it was a mix of cling and sea dance. Cling and sea dance. Boy, I haven't I can't remember the last time I had a great experience with Cling. I feel bad for them. Like they're okay. They are. So it was done long back. It was done like six to eight months back. We had very less options at that time. [...] of this and this just feels a lot less usable for me. So that's my long-winded answer there. our sundar. Um, so yeah, I I I don't feel like the the tagging and character references and stuff like