Runway announced that ElevenLabs v4 is now available on its platform; the model offers stronger expressiveness and more natural intonation for narration and character dialogue. v4 is available starting today alongside Runway's image and video models, and the official trial link is provided.
The Xiantu Qwen Image 2.1 Workbench (a ComfyUI creative project) fixed the Registry install package, excluding the .ruff_cache generated by the validation tool. Node functionality, interfaces, and workflow compatibility all remain unchanged.
ReelMimic is an open-source project; users provide a reference video and a one-sentence request, and it breaks down editing rhythm, shot length, transitions, composition, color, and camera movement, first generating a storyboard and style-frame plan for confirmation, then handing it to up to 6 AI agents for parallel production, with each shot reviewed by another agent, and fixes requiring before-and-after comparison screenshots.
Why it matters: A locally run multi-agent animation tool that breaks down a reference video's rhythm and camera movement into a reviewable creative plan.
Director's Console 0.4.10 adds hand-drawn motion paths, supporting ground, stairs, and object surfaces, with multi-stroke continuation and undo while retaining point selection. This version optimizes dragging and timeline editing of large imported models, moves OBJ parsing, packing, and validation into a background Worker, and adds curve adjustment and range sampling for depth frames.
mlx-audio released v0.5.7, submitted by lucasnewman, with 56 commits on the main branch since the previous version. This update bumps the version number to 0.5.7 and removes a redundant soundfile reference.
Second Unit released version 2026.09.28, updating the Windows download to the main branch and incorporating the workflow directories and fixes added after the September 15 package. The update includes MiniMax H3, LTX 2.5, SAM 3.1, WAN 2.2, and Juggernaut workflow directories, as well as model-specific prompt extensions, environment presets, seed handling, and an H3 Pose Follow fix.
Jimeng AI's web version launched a sample mode: first generate a draft at 480P, and after you are satisfied, natively upscale it to a 1080P final cut; the cost of 480P is about 1/8 of 1080P. The author tested a 15s video: two 480P generation attempts plus upscaling totaled 99 yuan, while generating directly at 1080P cost 156 yuan, saving about 38%; the final cut basically reproduces the sample, with no obvious difference to the naked eye in positioning, dialogue, core props, depth of field, or sound effects. Sample mode is also available in the canvas and retains Seedance 2.5 capabilities such as all-purpose reference, first and last frames, and smart editing, but does not yet support 3-minute ultra-long videos.
Why it matters: Jimeng launches sample mode, replacing direct high-definition generation attempts with 480P drafts plus native upscaling, which creators can use to recalculate storyboard preview and trial-and-error costs.
HelloGitHub Issue 126 features 31 open-source projects, including ccstatusline for customizing the Claude Code status bar, the hand-drawn-style image generation skill pack handraw-style that can be installed into Codex, and srt-whiteboard-animation, which turns SRT subtitles into whiteboard hand-drawn videos.
VideoLingo v3.1.1 is released, adding a Fish Audio official dubbing provider that uses the s2.1-pro model by default and can clone a voice from 15–30 seconds of speech in the original video without saving the voice in the account.
InvokeAI releases v6.14.2, fixing multiple bugs from 6.14.1 and adding two features: the Canvas text tool now supports importing custom fonts, and image style can be transferred to krea-2 via a reference image. This version also fixes issues with password reset tokens, path traversal protection, img2img posterization, and release of the quantized Qwen2.5-VL encoder in VRAM, and restricts Z-Image to using only the Qwen3 4B encoder.
Jimeng AI has launched "Sample Mode": first use 480P for low-cost draft pulls, then once a satisfactory take is selected, natively upscale it to a 1080P final cut. The author measured that a 30s video at 480P costs 270 credits per run, while direct 1080P output requires 1920 credits; 5 draft pulls plus upscaling totals 3270 credits, saving nearly 66% versus 9600 credits for all-direct 1080P output.
drama-skills has released v0.8.0, adding a "payoff-first" adaptation workflow: first rank a payoff table by audience benefit and then split into episodes, with the Episode 1 entry comparing at least three candidates, one of which is a cold open. This version adds three genre cards—system flow, rebirth/time travel, and comeback—writes camera movement shot by shot in the storyboard, adds on-screen text and sound effect rows to the edit sheet which render overlays onto the final cut, and supports the Wan 3.0 video prompt dialect (target_video_model: wan-3.0).
VideoLingo v3.1.0 has been released; the default local speech recognition engine is changed to Qwen3-ASR, and Qwen3-ForcedAligner is used to align word timings, with the 1.7B model selected by default and the 0.6B model also switchable.
SunoMV has launched a "one-sentence MV" feature: users write one sentence containing characters, place, genre, and a twist, and receive a song title, Suno genre tags, full lyrics, and four to six shot plans, with the plans drafted by Claude Opus 5.5.
The project panel of ComfyUI H3 Project Suite can now run inside SwarmUI; because the SwarmUI proxy discards request headers carrying the panel's security token, the token now falls back to the request body, and in file upload scenarios it falls back to the URL. Regular ComfyUI is unaffected; see SECURITY.md and CHANGELOG.md for details.
A developer has released a ComfyUI custom node, ForgeNeo Bridge, which lets you drag an image generated by Forge Neo into the ComfyUI canvas to read its model, prompt, seed, and dimensions and automatically lay out a workflow.
The open-source project killkli/minimax-h3-colab-skill is a standalone Codex skill that generates short MiniMax H3 Ref2VA videos from local reference images via Google Colab.
LibTV 1.5 has launched, with the Agent upgraded to TV Director and five new capabilities: original and adapted scripts, a character styling room, director's direction, 3D-BOX, and creative opening titles. The author tested it with a period short drama: TV Director can plan a 50-episode outline and per-episode scripts in one go, and scripts support manual editing like a document or box-selecting to hand to the Agent for revision; the character styling room can sculpt faces to adjust features, hairstyles, body types, and clothing, and batch-replace related assets on the canvas.
Why it matters: The author tests LibTV 1.5's script, character, storyboard and 3D previsualization workflow with a complete period-costume short drama, so readers can judge whether this workflow can replace existing production stages.
pyvideotrans releases v4.14 Windows pre-packaged version; the full package is 2.7G, and the 360MB patch package is only for overwriting the old package. This version optimizes speech, subtitle and image alignment algorithms, checks before Google Translate and automatically switches to Microsoft Translator when blocked by 429, and optimizes openai-whisper model downloads. MacOS/Linux require source deployment; if the new version has issues, users can continue using v4.13/v4.03.
Introducing the Pika API Club member spotlight—showcasing the developers and small teams building genuinely exciting work with the Pika API.
AI-Bridge @doerstokyo342, an AI studio based in Tokyo, offers video production, corporate training and custom AI tools. After joining the Pika API Club, their generative media costs were cut in half. Read the full story 👇
Tagtelier v0.5.18 adds a "sort order" for LoRA images, which can group display by folder and remembers the selection, and fixes the issue where LoRAs with the same name in different folders were merged and only one was shown; duplicate names are now displayed as "folder name/file name." This version also adds a LoRA commercial-use flag (which can be fetched in bulk from Civitai via the "Fetch commercial info" button), a ★ on favorite LoRA thumbnails, prescribed prompts registered by checkpoint or UNET, and support for negative LoRA strengths.
The tensor shape mismatch when using Qwen-Image 2.1 together with the TaylorSeer cache has been fixed and merged into the Diffusers main branch, in under four days from the issue being filed on the evening of September 21 to the merge on September 25.
The ComfyUI batch node plugin releases v0.2.0, adding the BatchLoadRecords general record iteration node, which can consume BATCH_RECORDS lists and work with any BatchSave* node, and supports template tokens such as {index} and {name} as well as custom field formats.
Claude Opus 5.5 meets Runway MCP.
Connect Runway directly to Claude to generate polished images and videos with Gen-4.5, Seedance 2.5, GPT Image 2, Kling and more, right in your workspace. Connect in seconds via the link below.
explainroo is a free open-source tool that lets coding agents generate explainer videos from a topic, with voiceover, timing and rendering all done on the local machine and the final output as an MP4. The agent only needs to write two files, script.md and scenes.js; Kokoro handles narration (28 voices, no account or API key needed), Whisper marks the pronunciation time of each word, Chrome draws the visuals in the background, and ffmpeg composites the video.
The open-source project muse2api reverse-engineers and wraps the multimodal capabilities of the muse.ai web app into an OpenAI-compatible RESTful API, supporting /v1/chat/completions, /v1/responses, text-to-image, image-to-image, text-to-video and first-frame image-to-video.
Runway introduces a new feature, Layers, which can split any image into editable layers in one click. Users can remove backgrounds, modify text and adjust any element individually within Runway, without leaving Runway.
ComfyUI-Easy-Media released v1.3.3, adding a multi-image loader and fixing an audio lockup issue caused by video audio extraction in multi-track projects. This version adds task image muting, task clip merging, and batch prompt application to the multi-track editor, supports frame-accurate pause preview, and allows SelfLift's high-resolution stage to use a separate model.
Promptsref's September 25 issue selected this week's top three Midjourney SREFs: cinematic anime hybrid, candy-colored 3D animation hybrid, and graphic cartoon single-code style. All three center on character design, suited respectively to visual novel scenes, mascot concepts, and avatar design. The official note clarifies these are only creative direction references, not actual outputs.
ChatGPT's /leeveo command can transform the same character into different faces, ages, outfits, expressions, poses, and story worlds while maintaining identity consistency. The official demo uses Gandalf, Frodo, Mad Max, and The Bride as references, with one reference character branching into ten creative directions for character building in AI films, storyboards, games, marketing campaigns, and visual narratives.
Your Grok Bots are now production studios. Connect them to the Pika API, and they can generate images, videos, and audio using 120+ models through a single API key: Seedance 2.5, Wan 3.0, GPT-image-2, and more.
ComfyUI-PerfectVideoResolution released a workflow update, rewriting the changelog extraction logic in pure shell. Previously, the embedded Python (pathlib/subprocess) in the workflow was flagged by the registry security scanner as python_sensitive_file_access, causing version 0.1.1 to be blocked.
Tern is a podcast and video library search tool that runs locally on Apple Silicon, using three-way indexing via whisper.cpp (Whisper Large v3 Turbo, Q5), Apple Vision OCR, and SigLIP-2 to retrieve any clip by voice, on-screen text, or visual content, with results merged into a unified ranking.
Runway launched Draft mode for Seedance 2.5, which generates faster and consumes fewer credits during the exploration phase, allowing users to later enhance satisfactory results to full quality.
BridgeMind released the open-source desktop app BridgeClip, which can find complete clips from podcasts, livestreams, and interviews, then edit and export subtitled vertical 9:16 or horizontal 16:9 short videos. Video rendering is done locally, AI processing uses the user's own OpenRouter key, and it supports macOS, Windows, and Linux, with the installer bundling Python, FFmpeg, and yt-dlp.
Why it matters: The open-source desktop tool chains long-video clipping, subtitling, and publishing into a single local rendering pipeline, enabling creators to evaluate how to repurpose their own assets.
OpenCut released a pre-release of FFmpeg 8.1.3 (build 1), providing an LGPL shared build for its media crate, compiled by the Media dependencies workflow. Each archive includes the exact configure parameters (BUILD.md) and the FFmpeg license, and the version can be pinned via bun run crates/media/setup/pin.ts ffmpeg-8.1.3-1.
Trendlume v0.3.0 was released, upgrading the video generation engine from a single text pipeline to a three-track production system covering knowledge science popularization, e-commerce product promotion, and story short dramas.
SmartSub released v3.9.0, adding a built-in AI creation assistant (shortcut ⌘J / Ctrl+J) that supports context awareness, conversational subtitle polishing, screenshot troubleshooting, and file dialogue. This version supports the MCP protocol and CLI automation, providing 111 tools and a unified command line, with one-click connections to Cursor, Codex, and Claude Code, and a new audio-video toolbox covering audio extraction, subtitle merging/conversion/alignment, video compression, trimming, and GIF conversion.