GenBench Drops Audio Research, Seedance 2.5 Hits Full Capacity, and the AI Film Fest Is Coming to LA
Three weeks of momentum packed into one read. The AI filmmaking stack moved fast this October, and Genflick's research desk was first to measure it. Here's what happened.
GenBench: 97 Reports and a Full Audio Stack Measured
Genflick's research desk at genflick.com/genbench crossed 97 published benchmark reports this week. The freshest batch (Oct 2-3) is all audio: ElevenLabs Music v2.5 cost and speed across 26 songs, speech-to-text accuracy for song lyrics, and a deep dive into the steady tone sitting under Seedance audio that nobody was talking about.
The Seedance audio study is the one to read. GenBench measured 27 takes across Seedance 2.0 Mini, Fast, Regular, and 2.5, found a persistent steady tone under the ambient layer, and tested an adaptive filter that removed it without killing the surrounding sound. If you're rendering dialogue scenes with Seedance, the cleanup path is documented.
They also benchmarked Gemini 3.8 Flash labeling male and female voices in songs against ElevenLabs speaker separation across 27 real songs and 10 duets with known answers. The Flash model won on cost and speed. Full results and audio samples are on GenBench.
97 reports. The research desk is doing the work nobody else is doing in this space. Go read it.
Seedance 2.5 Is Running at Full Capacity on Genflick
Seedance 2.5 (Video phase, Genflick-supported) is now at full production capacity. The model generates 30-second native clips in a single pass at up to 4K resolution with synchronized audio. The reference system expanded to 50 multimodal inputs (up to 30 images, 10 video clips, 10 audio clips in one generation) with clay render referencing for spatial structure blocking, motion referencing for camera paths, and improved lighting control.
The practical shift: you can now plan scenes that last. The 10-second clip era is over. A single Seedance 2.5 generation on Genflick covers a full scene setup, motion progression, and resolution without stitching fragments frame by frame. The GenBench audio research confirms the synced sound is usable after the adaptive filter pass.
GPT Image 2.5 Sunburst: The New Image-Phase Default on Genflick
OpenAI replaced the single GPT Image 2 with two variants on Sep 8: GPT Image 2.5 Flare (faster, lighter) and GPT Image 2.5 Sunburst (heavier, built for detailed creative work with longer generation times). Sunburst is the current default in Genflick's Image phase.
For AI filmmakers: Sunburst handles character reference sheets, set designs, and keyframe images with higher fidelity than Flare. Flare is your rapid-iteration model. Use Sunburst when the frame matters. Use Flare when you're testing compositions. Both are selectable in Genflick's Film Studio.
Runway Agent Adds Seedance 2.5 Draft Mode
Runway's Agent can now generate with Seedance 2.5 Draft mode, creating fast 480p drafts with post-generation enhancement. Industry context only. The takeaway for Genflick users: Seedance 2.5 is a Genflick-supported Video phase model, and the Draft-mode approach validates the same iterate-cheap-then-commit-premium workflow Genflick's three-phase pipeline already enables. Generate cheap, evaluate, commit the keeper.
The Genflick AI Film Fest Is Coming to LA Tech Week
Oct 13. 7 PM. LA Tech Week. The Genflick AI Film Fest is a film hackathon. Bring your series, your script, or just an idea. Present a 3-5 minute treatment to the room. We're building the AI filmmaking stack and the community that uses it.
If you've been building something on Genflick or testing the pipeline, bring it. The room is the room.
The Stack Right Now
Genflick runs a three-phase Film Studio pipeline: AI Director (story development, planning, prompt authoring) and Image / Keyframe generation (character references, sets, props, storyboard images) into Video generation (prompts and visual references into moving clips). The models that matter this week:
AI Director: Meta Muse Spark 1.3 (default), Grok 4.7, GPT-6 Astra, Claude Opus 5.5, Gemini 3.1 Pro, Gemini 3.8 Flash, Groq Pro, Groq Max.
Image: GPT Image 2.5 Sunburst (default), GPT Image 2.5 Flare, Grok Imagine, Nano Banana Pro, Nano Banana 2.
Video: Seedance 2.5, Seedance 2.0 Mini/Fast/Regular, MiniMax H3, H3 Max, Grok Imagine Video, Veo 3.1, Kling 3.0/O3, Wan 2.6, Runway Gen-4.
Not every model every edition. This week the momentum is Seedance 2.5 at full capacity, GPT Image 2.5 Sunburst as the new Image default, and GenBench's audio research. Next week, the Film Fest.
Genflick is building the AI filmmaking stack. Sign up: genflick.com
Research cited from GenBench (97 published reports). Model catalog from genflick.com.