Research: ai video editing and creating - comprehensive landscape May 2026¶

URL: https://mkdocs.justinsforge.com/memory/research/ai-video-editing-and-creating-landscape-2026-05-02/

Date: 2026-05-02 Depth: deep Model: sonnet

TL;DR¶

Hyperframes (HeyGen, Apache 2.0, 13.8k stars, v0.4.42 shipped 2026-05-02) is the clear winner of the agentic-renderer battle: HTML-in, MP4-out, no per-render fees, native Claude Code skill, deterministic output, and the most active community trajectory.
The official Adobe-Claude connector (April 28, 2026) exposes 50+ tools across eight Creative Cloud apps including Premiere but Premiere-specific tool granularity is undisclosed; it is Claude.ai-web only for now, not Claude Code CLI, making it awkward for forge pipeline use.
DaVinci Resolve has the richest MCP ecosystem: samuelgursky (354 tools, MIT, 945 stars, v2.3.0 April 29 2026) is the mature baseline; CutMaster AI (233 tools, 0 stars, nascent) is more ambitious but unproven; both require Resolve Studio ($295 one-time).
For transcription in a headless pipeline, whisper.cpp (free, local, no API spend) is the forge-correct choice for batch work; ElevenLabs Scribe v2 wins on real-time accuracy (3.5% WER, 150ms) but costs API money.
B-roll video generation is all pay-per-second API spend: no model in this space is free or open-weight at quality parity; Kling 3.0 via fal.ai (~$0.029/sec) is the cheapest credible option; open-weight Wan 2.2 is the only zero-cost path with acceptable output.

Findings¶

1. Agentic / programmatic renderers¶

Hyperframes (HeyGen, Apache 2.0) Released April 17, 2026. Architecture: AI agents write plain HTML/CSS/JS, Hyperframes captures frames via headless Puppeteer, encodes with FFmpeg, outputs MP4 [1][2]. Ships with a --non-interactive flag for pipeline use, a built-in MCP-compatible skill directory, and native Claude Code/Cursor/Gemini CLI/Codex support via npx skills add heygen-com/hyperframes. Catalog includes 50+ pre-built blocks (GSAP, Lottie, Three.js, Anime.js). 13,800 stars, 1,200 forks, v0.4.42 shipped same day as this report (2026-05-02) [2]. License: Apache 2.0, fully commercial, no seat caps, no per-render fees. Render speed not publicly benchmarked but is Puppeteer-frame-capture speed, which is CPU/GPU dependent on the host. LLM-friendliness: highest in class because compositions are vanilla HTML that every model already knows how to write [3].

Remotion (source-available, paid above 3 people) React-based programmatic renderer, well-established (predates Hyperframes by ~3 years). Powerful for data-driven React compositions. License: Remotion License (source-available); free for individuals and companies under 4 employees; paid above that at $25/seat/month minimum $100/month company license [4]. No per-render fee. Render is Node.js + headless Chromium + FFmpeg, same stack as Hyperframes essentially. LLM-friendliness: moderate; React JSX is harder for LLMs to compose correctly than plain HTML. Ecosystem: mature, extensive docs, used in production by many teams. Used inside OpenMontage as one of its two rendering engines.

Rendervid (QualityUnit, FlowHunt Attribution License) JSON-template renderer, MCP-native with 6 exposed tools (render_video, render_image, validate_template, get_capabilities, list_examples, get_example) [5]. 31 stars, 6 forks -- very early. License is "FlowHunt Attribution License" (attribution required, not OSI-approved). Render approach: headless browser + FFmpeg (server-side) or browser renderer. Supports 40+ animation presets and custom React components. Lowest adoption of any named tool in this category. License is the weakest link for commercial forge use.

OpenMontage (AGPLv3, 3.3k stars) Not a renderer per se -- it is an agentic video production orchestrator. Sits above renderers: it uses Remotion and Hyperframes as downstream render engines while itself handling research, scripting, voice-over (Piper TTS for zero-API path), asset generation, and final composition [6]. 12 pipelines, 52 tools, 400+ agent skills, 665 forks. AGPLv3 license: copyleft, technically free but any service built on it must open-source its modifications. Launched/featured April 24, 2026. The zero-key path (Piper TTS + Archive.org/NASA footage + local FFmpeg + Remotion) aligns well with forge's subscription-quota-only preference.

IMG.LY CE.SDK Enterprise-grade SDK targeting apps that embed an editing surface (mobile, web, Electron) rather than headless pipelines. Has a server-side GPU-accelerated renderer for batch production [7]. Pricing: not public, requires sales contact. No fixed per-render fee visible. Not meaningfully LLM-friendly; no MCP server found. Poor fit for Justin's use case: it is an SDK for building user-facing editors, not an agent-driven pipeline tool.

Twick (MIT, GitHub) React-based video editor SDK with canvas timeline, drag-and-drop, AI captions, and serverless MP4 export [8]. Intended for developers building video app UIs (similar positioning to IMG.LY but lighter weight). Not a headless pipeline tool. No MCP layer.

Who is winning the agentic-video battle: Hyperframes, by a significant margin. 13.8k stars versus Remotion's 22k (but Remotion is 3 years older), OpenMontage's 3.3k, and everyone else in the single-digit hundreds. Hyperframes has the Apache 2.0 license, the easiest LLM interface (HTML), and the most ecosystem momentum as of May 2026 [1][2][3].

2. AI trim / transcribe / caption / clean-up tools¶

video-use (browser-use team, ~MIT, 6k stars) The most Claude-Code-native tool in this category [9]. It does not require the LLM to watch video frames; instead it builds a structured transcript + visual composite (12KB text + key PNGs) for LLM reasoning. Uses ElevenLabs Scribe for word-level timestamps and speaker diarization. Generates animation overlays via Manim, Remotion, or PIL in parallel sub-agents. Includes self-evaluation loops at each cut boundary. Dependencies: Python, FFmpeg, ElevenLabs API (cost), yt-dlp. Strong fit for forge pipelines but requires ElevenLabs API spend.

Descript (SaaS, API beta open) Transcript-based text editing (delete words in transcript, video updates). API is now in open beta for programmatic project creation, import, and editing at scale [10]. Underlord AI assistant now runs on Claude Opus 4.7 internally. Pricing: SaaS subscription (~$12-24/month creator tier) plus API costs (enterprise). Not self-hosted; data leaves your machine. Poor forge fit on sovereignty grounds unless Justin already pays for it.

whisper.cpp (MIT, local) MIT-licensed C++ port of OpenAI Whisper. Runs fully local, zero API spend, GPU-accelerated. Processes ~1 hour of audio in 6-18 minutes on GPU [11]. WER approximately 10.6% in real-world conditions (Whisper Large V3 Turbo). Best-in-class for forge: free, local, scriptable from Claude Code, no per-token spend. Ideal for batch transcription in pipelines.

OpenAI Whisper API $0.006/min, WER ~5.3% [11]. Costs API money (incompatible with Justin's subscription-quota-only stance unless routed through n8n). GPT-4o-Transcribe at same price point achieves 4.1% WER and is now the cost-performance leader for cloud transcription. Skip unless whisper.cpp accuracy is insufficient.

ElevenLabs Scribe v2 150ms real-time latency, WER ~3.5%, speaker diarization, 90+ languages [12]. Best accuracy of any option. However it costs API money (not subscription quota). Worth it for final-pass caption burns but not for bulk rough-cut pipeline work. The video-use tool uses it, so if Justin adopts video-use, Scribe spend is implicit.

AssemblyAI $0.0043-$0.0077/min batch/streaming, WER ~6.84% [11]. Cloud SaaS, no local option. Adds content moderation, sentiment, topic detection. Overkill for Justin's use case; Scribe or whisper.cpp both outperform it on accuracy at similar or lower cost.

AutoEditor (WyattBlue, MIT, CLI) Command-line tool for automated silence removal via audio loudness analysis. Supports silence-detect, silence-removal, silence-speedup, motion-detection [13]. Pure local Python CLI. Perfect for automated rough-cut prep before LLM-driven editing. Zero cost, fully scriptable. Pairs well with whisper.cpp: AutoEditor removes silences, whisper.cpp transcribes, Claude edits by transcript.

Pipeline recommendation for forge: AutoEditor (silence removal) > whisper.cpp (transcription) > Claude Code (edit-by-transcript reasoning) > Hyperframes or Remotion (re-render) > FFmpeg (final encode). No API spend except optional Scribe for final caption accuracy.

3. AI hooks into legacy NLEs¶

Official Adobe-Claude connector (April 28, 2026) Anthropic + Adobe shipped the "Adobe for creativity" connector as part of a nine-connector "Claude for Creative Work" release [14][15]. Exposes 50+ tools across Photoshop, Illustrator, Firefly, Express, Premiere, Lightroom, InDesign, and Stock. Claude orchestrates multi-app workflows from a single prompt. Critical limitation: access is through claude.ai web/desktop/mobile Connectors panel -- not the Claude Code CLI. This means it cannot be invoked from a terminal script or forge pipeline without a UI session. For Justin's eval round, this works only in a Claude.ai browser session on Sol (Mac) or Console via a browser, not headlessly. Premiere-specific tool inventory is not publicly disclosed in the launch materials; the only confirmed Premiere action is "resize and repurpose videos for social platforms" (crop/resize to platform dimensions) [16]. The full tool list requires consulting unpublished developer docs.

Community Premiere MCP servers - hetpatel-11/Adobe_Premiere_Pro_MCP: Covers project ops, ingest, sequence creation, timeline editing, transitions, effects, keyframes, metadata, exports, and higher-level assembly [17]. Uses ExtendScript + UXP plugin bridge. Works with Claude Code, Claude Desktop, any MCP client. Not endorsed by Adobe. - mikechambers/adb-mcp: Proof-of-concept by a former Adobe engineer. Controls Photoshop, Premiere Pro, InDesign via MCP. Self-described proof-of-concept, not production-hardened [18]. MIT license. - viaSocket: Cloud-hosted MCP relay that wraps Premiere actions for ChatGPT/Claude/Cursor. Adds a cloud dependency. Less useful for a sovereign forge setup. - toonyai/premiere-mcp: Listed on Lobehub but minimal public documentation found.

Verdict on community Premiere MCPs: The hetpatel-11 server has the broadest documented tool coverage and works with Claude Code CLI directly. Mike Chambers' adb-mcp is more trustworthy in pedigree (ex-Adobe engineer) but explicitly self-described as a proof of concept. Neither has large star counts or production track records. Real-world reliability for complex timeline edits is unverified.

DaVinci Resolve MCP servers - samuelgursky/davinci-resolve-mcp: The mature choice. MIT, 945 stars, v2.3.0 shipped April 29, 2026 [19]. 354 tools (granular) or 27 tools (compound default). Covers 98.5% of the Resolve Scripting API. Actions span timeline manipulation, color grading, Fusion compositing, media pool, markers, exports (EDL/XML/AAF), and render presets. Requires DaVinci Resolve Studio (free version has no external scripting API). Studio is $295 one-time, no subscription. - barckley75/resolve-mcp: Smaller, simpler MCP server for Resolve Studio. Fewer tools than samuelgursky. MIT. - CelaviiHQ/cutmaster-ai: 233 tools (203 granular + 14 workflow + 10 AI via Gemini + 5 resources + 9 skills/7 agents) [20]. MIT. Also requires Resolve Studio. Optional Gemini API key for AI features. 0 stars, 158 commits: very new and unproven in production. Ambitious but nascent. - apvlv/davinci-resolve-mcp: Separate Resolve + Fusion-specific implementation, also on GitHub.

Final Cut Pro Multiple community MCP servers now exist: fcpxml-mcp-server (DareDev256), elliotttate/finalcutpro-mcp (88 tools, FCPXML engine + live FCP control, released April 23 2026), and FCPBridge (modding platform for direct FCPX access) [21]. Apple Creator Studio launched with AI beat detection, search, and montage automation powered by Apple Intelligence. FCP is macOS-only which means Sol only, not Console. Maturity of MCP servers is unknown but early signals are promising.

4. AI video / image generators for B-roll¶

Sora 2 (OpenAI) Best physics realism. Official API exists at ~$0.10/sec [22]. Also accessible via ChatGPT Plus ($20/month) or Pro ($200/month) subscriptions -- but web UI only, not programmable. Third-party aggregators (AI/ML API) offer ~$0.08-$0.12/sec. Programmatic API spend is incompatible with Justin's subscription-quota-only stance. Web-UI access via ChatGPT Plus subscription is technically $0 marginal cost but not automatable. Rated best for documentary/B-roll realism [23].

Veo 3.1 (Google DeepMind) Highest cinematic quality tier. Official API on Vertex AI at $0.75/sec standard, $0.05/sec for 720p Lite (launched March 31, 2026) [22][23]. Also available through Runway API (Runway added Google Veo3 as a third-party model on its API in September 2025). Per-second pricing means API spend; not subscription-quota.

Kling 3.0 (Kuaishou) Cheapest credible option at ~$0.029/sec via fal.ai or Replicate [22]. Native 4K output. No official public API from Kuaishou as of this research; third-party aggregators only. Entry-level paid plan $10/month on kling.ai with commercial rights. Motion control is a standout feature (reference video transfer).

Runway Gen-4.5 Official Runway REST API, available since February 10, 2026 [24]. Text-to-video and image-to-video, 2-10 second clips. Subscription pricing ($12-$76/month plans) rather than pure per-second. Ranked #1 on Artificial Analysis Text-to-Video Benchmark at launch (1,247 Elo points). Nano Banana (Gemini 2.5 Flash video) is available via the Runway API as a third-party model since September 2025 [24]. Runway subscription included in plan means lower marginal cost than per-second APIs if Justin already subscribes.

Seedance 2.0 (ByteDance) Proprietary, closed-source. API launched April 2, 2026. Accepts text, image, video, and audio inputs in one request [25]. Pricing via aggregators. Not open source. Best overall motion + audio sync per some rankings [23].

Wan 2.2 (open-weight) The only credible open-weight option for local generation. Runs via ComfyUI. Quality is competitive but below Sora 2 / Veo 3.1. Zero API cost; runs on Justin's own GPU. Forge-doctrine-aligned if Justin has adequate GPU. Console (Linux VM) GPU availability would need verification.

Nano Banana Available via Runway API as a hosted model (Gemini 2.5 Flash variant). Also available standalone at nanobananas.ai. Runway Academy published Gen-4.5 + Nano Banana Pro tutorials for "cinematic universe" creation [24]. API access via Runway subscription or per-credit.

B-roll strategy for Justin: If budget is a concern, Wan 2.2 locally (zero cost) is the forge-doctrine play. If quality matters more, Kling 3.0 via fal.ai (~$0.029/sec) is the cheapest commercial option. Runway Gen-4.5 via subscription amortizes cost better than pure per-second models if already subscribed. None of these fit "subscription quota only" cleanly since they all charge per-second or per-credit separately from Claude subscription.

5. New entrants since April 2026 (bonus)¶

Claude for Creative Work connector suite (April 28, 2026): Beyond Adobe, Anthropic shipped eight other connectors simultaneously: Blender, Autodesk Fusion 360, Ableton, Splice, Affinity by Canva, SketchUp, Resolume, and more [14]. The Blender connector is particularly notable for 3D asset creation for video work.

Apple Creator Studio: Apple announced an AI subscription tier for Final Cut Pro with beat detection, AI-powered search, and automated montage assembly [21]. macOS-only, Apple Intelligence-powered. Not yet shipping as of this research date.

FCPBridge: A modding platform for FCP X that gives LLMs direct programmatic access to the timeline (not just accessibility features) [21]. Very early, macOS-only.

VideoAgent (HKUDS): Academic-origin agentic framework for video understanding, editing, and remaking. Decomposes instructions into sub-intents and maps to a multi-agent capability graph. GitHub: HKUDS/VideoAgent. Research-grade, not production-hardened.

ViMax (HKUDS): Combines Director, Screenwriter, Producer, and Video Generator agents into one agentic video generation system. Also academic/research origin.

Open-Sora 2.0 (hpcaitech): 24.1k stars, the most-starred open-source video generation model on GitHub. Runs locally, free weights. Alternative to Wan 2.2 for local B-roll generation.

Comparison matrix¶

Tool	Category	License	Render speed	LLM-friendliness	Maturity	Forge fit
Hyperframes	Agentic renderer	Apache 2.0	Puppeteer+FFmpeg, no published benchmark	Highest (plain HTML)	High (13.8k stars, v0.4.42, May 2026)	Excellent
Remotion	Agentic renderer	Remotion License (paid above 3 people)	Node+Chromium+FFmpeg	Moderate (React JSX)	Very high (3+ years, 22k stars)	Good (license friction)
Rendervid	Agentic renderer	FlowHunt Attribution (non-OSI)	Headless browser+FFmpeg	High (JSON templates)	Very early (31 stars)	Poor (license + maturity)
OpenMontage	Agentic orchestrator	AGPLv3	Delegates to Remotion/Hyperframes	Very high (YAML+Markdown skills)	Early (3.3k stars, April 2026)	Good (copyleft risk)
IMG.LY CE.SDK	Embedded editor SDK	Commercial (quote only)	GPU server render	Low (no MCP)	High (enterprise)	Poor (cost, no CLI fit)
Twick	Editor SDK	MIT	Serverless MP4	Low (no MCP)	Very early	Poor (not pipeline tool)
video-use	Trim/transcribe/edit	MIT (unconfirmed)	FFmpeg-native	Very high (Claude Code skill)	Early (6k stars)	Good (ElevenLabs cost)
Descript	Trim/transcribe	SaaS (proprietary)	Cloud	Low (GUI-first, API beta)	High (production SaaS)	Poor (SaaS, data sovereignty)
whisper.cpp	Transcription	MIT	1hr audio in 6-18min GPU	High (CLI scriptable)	Very high (years, huge community)	Excellent
AutoEditor	Silence removal	MIT	Real-time-ish FFmpeg	High (CLI scriptable)	High (active PyPI)	Excellent
ElevenLabs Scribe v2	Transcription	Proprietary API	150ms real-time	High (API)	High	Moderate (API spend)
AssemblyAI	Transcription	Proprietary API	Batch/stream	Moderate (API)	High	Poor (cost vs whisper.cpp)
Adobe connector (official)	NLE hook	CC subscription required	N/A (cloud)	Low (web UI only, no CLI)	New (April 28, 2026)	Poor for headless pipeline
hetpatel-11 Premiere MCP	NLE hook	Not specified	N/A (Premiere)	High (Claude Code MCP)	Early (community)	Moderate (unproven reliability)
mikechambers adb-mcp	NLE hook	MIT (proof-of-concept)	N/A (Premiere)	High (Claude Code MCP)	Early (PoC)	Moderate
samuelgursky Resolve MCP	NLE hook	MIT	N/A (Resolve)	Very high (354 tools)	High (945 stars, active)	Very good (Resolve Studio cost)
CutMaster AI (CelaviiHQ)	NLE hook	MIT	N/A (Resolve)	Very high (233 tools+agents)	Very early (0 stars)	Moderate (unproven)
FCP MCP (elliotttate, 88 tools)	NLE hook	Unknown	N/A (FCP)	High	Early (April 2026)	Moderate (macOS/Sol only)
Sora 2	B-roll generator	Proprietary API	Variable	Low (API spend)	High	Poor (API cost)
Veo 3.1	B-roll generator	Proprietary API	$0.05-$0.75/sec	Low (API spend)	High	Poor (API cost)
Kling 3.0	B-roll generator	Proprietary API	~$0.029/sec	Low (API spend)	High	Moderate (cheapest)
Runway Gen-4.5	B-roll generator	Subscription + credits	2-10s clips	Moderate (REST API)	High	Moderate (subscription)
Wan 2.2	B-roll generator	Apache 2.0	Local GPU	Moderate (ComfyUI)	Moderate	Good (zero cost, GPU needed)
Open-Sora 2.0	B-roll generator	Apache 2.0	Local GPU	Moderate	High (24.1k stars)	Good (zero cost, GPU needed)

Disagreements / open questions¶

Hyperframes render speed: No published benchmark. Multiple sources confirm the Puppeteer+FFmpeg architecture but zero timing data. Render speed is highly machine-dependent. This must be measured in Justin's actual eval, not assumed.
Official Adobe connector Premiere tool inventory: Adobe and Anthropic's public materials do not enumerate Premiere-specific MCP tools. The "50+ tools" figure covers all eight CC apps. Actual Premiere coverage could be 3 tools or 20; this is genuinely unknown without developer docs. The community hetpatel-11 server has a more documented (if less officially supported) Premiere tool set.
video-use license: The README does not state a license explicitly. The browser-use parent organization uses MIT for their main project; video-use is assumed MIT but this is unverified. Do not treat as confirmed for production commercial use without checking.
Kling 3.0 official API: Multiple sources say Kling 3.0 has no official public API and is third-party only (fal.ai, Replicate). One source implies a $10/month kling.ai plan. There is disagreement on whether direct Kuaishou API access exists. Treat third-party aggregator as the only reliable path for now.
Resolve Studio requirement: All three Resolve MCP servers (samuelgursky, barckley75, CutMaster AI) require Resolve Studio ($295 one-time). The free DaVinci Resolve version has no external scripting API. This is a prerequisite cost for any Resolve eval round.

Recommendation for Justin's eval¶

Prioritized round order:

Round 1: Hyperframes (clear first) Run it on Console (Linux, Node.js 22+, FFmpeg already present). Write a 60-second explainer-style video in HTML with GSAP animations as the test composition. This validates the zero-API-spend path, tests render speed on Console hardware, and exercises the Claude Code skill integration. Apache 2.0 license means no friction at any scale. Hyperframes is almost certainly the winner of the agentic-renderer battle; this round confirms it and gives you a performance baseline. Expected result: working, fully automated, no API spend.

Round 2: Remotion Same test composition ported to React. This measures the real cost of "LLM writes JSX instead of HTML" in terms of agent iteration loops and error rate. Also surfaces the license question concretely: with Justin solo, Remotion is free; if a contractor touches it, it costs $100/month minimum. Use OpenMontage's zero-key path (Piper TTS + Remotion render engine) to test the full orchestration layer if time allows. Expected result: slower agent iteration than Round 1 due to JSX complexity, but excellent final output quality.

Round 3: Premiere via community MCP (hetpatel-11) Install the hetpatel-11 MCP server, connect Claude Code on Console, and attempt a real edit task: import a 5-minute talking-head clip, remove marked silences, add a lower-third title card, export H.264. This tests the community server's actual reliability. If this works reliably, skip the official Adobe connector round (Round 4 below) since the CLI path is more valuable to forge than the web-UI path. Expected result: mixed; community MCPs for Premiere are unproven. Budget for debugging ExtendScript bridge issues.

Round 4: Premiere via official Adobe connector Only run this round if Round 3 is unreliable. Open a claude.ai browser session on Sol (Mac, has Creative Cloud). Test the same edit task. The official connector is more likely to be stable but is web-UI-only, not automatable from forge pipeline scripts. If Round 3 works, skip Round 4 entirely -- a web-UI-only connector that cannot be invoked from Claude Code CLI has no place in the forge stack.

Round 5 (optional): DaVinci Resolve via samuelgursky MCP Requires Resolve Studio ($295 one-time). If Justin does not already own Studio, this is the gating cost question. If he has it (or is willing to buy it for the eval), samuelgursky's server is the most mature NLE MCP in existence (945 stars, 98.5% API coverage, v2.3.0 active). Resolve's free version is a viable NLE for non-scripting tasks; Studio unlocks the scripting API for MCP use. Run the same 5-minute clip edit task. Expected result: most reliable NLE MCP path if Resolve Studio is available.

Skip these: - CutMaster AI: 0 stars, no production validation. Revisit in 3 months. - Rendervid: Non-OSI license, 31 stars. Not ready. - IMG.LY CE.SDK: Not a pipeline tool; requires sales call for pricing. - AssemblyAI: Costs API money, worse accuracy than whisper.cpp or Scribe. - Sora 2 / Veo 3.1 for B-roll: Per-second API costs conflict with doctrine. If B-roll generation is needed, test Wan 2.2 locally first; if quality is insufficient, Kling 3.0 via fal.ai is the lowest-cost commercial path.

Likely winner call: Hyperframes for agentic composition work (Round 1 almost certainly wins). For existing footage editing (NLE path), samuelgursky Resolve MCP is the most capable tool if Resolve Studio is available; hetpatel-11 Premiere MCP is the right test if Justin prefers staying in Premiere and already has Creative Cloud. The official Adobe connector is not a forge-pipeline tool in its current form.

Sources¶

GitHub: heygen-com/hyperframes, Apache 2.0 agentic HTML-to-MP4 renderer, v0.4.42 May 2, 2026
HeyGen HyperFrames Guide (antigravity.codes), architecture and agent integration overview
HeyGen HyperFrames: HTML to MP4 for AI Agents (creativeainews.com), LLM-friendliness analysis
Remotion License and Pricing, official pricing tiers, source-available terms
GitHub: QualityUnit/rendervid, JSON-template renderer, 6 MCP tools
GitHub: calesthio/OpenMontage, AGPLv3 agentic video production system, April 2026
IMG.LY CE.SDK Renderer announcement, server-side GPU render for automation
GitHub: ncounterspecialist/twick, React video editor SDK with serverless export
GitHub: browser-use/video-use, Claude Code agentic video editor, 6k stars
Descript API early access, programmatic project/edit API
Whisper vs ElevenLabs vs AssemblyAI comparison (deepgram.com), WER and pricing benchmarks 2026
ElevenLabs Scribe v2 150ms latency (tokenmix.ai), real-time STT specs
GitHub: WyattBlue/auto-editor, MIT silence/motion removal CLI
Adobe for creativity connector (blog.adobe.com), April 28, 2026 launch announcement
Claude for Creative Work (anthropic.com), nine-connector release overview
Claude + Adobe Creative Cloud: 50+ Tools (pasqualepillitteri.it), connector feature summary
GitHub: hetpatel-11/Adobe_Premiere_Pro_MCP, community Premiere MCP with ExtendScript bridge
GitHub: mikechambers/adb-mcp, ex-Adobe engineer proof-of-concept, MIT
GitHub: samuelgursky/davinci-resolve-mcp, MIT, 945 stars, 354 tools, v2.3.0 April 29, 2026
GitHub: CelaviiHQ/cutmaster-ai, 233-tool Resolve MCP + Gemini AI tools, MIT, nascent
Final Cut Pro AI and Apple Creator Studio (nofilmschool.com), FCP AI features and MCP servers
AI Video API Pricing 2026: Seedance vs Sora vs Kling vs Veo (devtk.ai), per-second pricing comparison
Best AI Video Generators 2026 (aimlapi.com), quality rankings and use-case routing
Runway Gen-4.5 and API changelog (runwayml.com), Gen-4.5 launch, Nano Banana and Veo3 as third-party models
Seedance 2.0 open source status (glbgpt.com), proprietary, API launched April 2, 2026

Search trail¶

Hyperframes HeyGen agentic video renderer Apache 2.0 2026
Remotion React video programmatic renderer license 2026
Adobe Claude connector Premiere Photoshop MCP tools April 2026
DaVinci Resolve MCP server samuelgursky davinci-resolve-mcp 2026
hetpatel-11 Adobe Premiere Pro MCP mikechambers adb-mcp community 2026
Sora 2 Veo 3 Kling video generator API pricing 2026
Rendervid OpenMontage IMGLYCE SDK Twick agentic video renderer 2026
video-use browser-use GitHub agentic video editor 2026
CelaviiHQ cutmaster-ai DaVinci Resolve MCP 227 tools barckley75 resolve-mcp 2026
Runway Gen-4.5 Nano Banana video generator API 2026
video-use Descript AutoEditor whisper.cpp ElevenLabs Scribe AssemblyAI pipeline Claude Code 2026
Final Cut Pro AI integration MCP 2026
whisper.cpp vs OpenAI Whisper API vs ElevenLabs Scribe accuracy speed cost comparison 2026
IMG.LY CE.SDK license pricing headless video render agent 2026
Descript video editor 2026 transcript-based editing programmatic API
Seedance ByteDance video model API 2026 Wan 2.6 open source
Adobe Premiere Pro MCP connector official vs community comparison what works 2026
agentic video rendering new entrants April May 2026 open source