The first story-based AI production platform. Create character-driven narratives with full creative control. Built for agencies, studios & filmmakers.
Boston, MA
Joined November 2021
- Tweets91
- Following506
- Followers359K
- Likes4.7K
Three witches throw a strip of 35mm film and a clapperboard into a cauldron. Watch what rises out of the pot.
Something big is brewing at @FilmSparkAI
Happy Halloween!!! 🎃
Review & Approval is live in FilmSpark. In a Team Workspace, shots and assets now carry a status from To Do to Approved, with Notes right beside it.
Also new: FLUX 3 video in 2K and 4K, and a new Images tutorial: youtu.be/I-_37rkzSig
2,947
All 19 FilmSpark tutorials on one page: filmspark.ai/tutorial-videos
New: a 2-min Scene Composer walkthrough.
Build shots and short sequences without a script.
youtu.be/cTdMJgmmNWk
Also shipped: MiniMax H3 Max 3D to Video for storyboard shots
Happy @FilmSparkAI Friday!
Seedance 2.5 Draft is live on @FilmSparkAI
Generate at 480p, re-roll until the motion is right, then regenerate the approved shot at 1080p.
Also new: GPT 6 Sol, GPT 6 Luna, and Claude Opus 5.5 for the Agent.
filmspark.ai
@FilmSparkAI has been selected to join the NVIDIA Inception Program!
Inception is NVIDIA's program for the AI startups it feels provide real value. FilmSpark is the production console for AI filmmaking: script to storyboard to character-consistent video, 30+ models, one workspace, built for studios.
Big milestone for the team. More soon!
#NVIDIAInception @NVIDIAAI
Create incredible Camera Moves!
@FilmSparkAI added #MinimaxH3MaxMultiAngle this week. Set keyframes and use the camera orbit controls to generate a multi-angle video, now with 1080p output. The demo is below.
Also new this week:
• GPT Image 2.5 Flare and Sunburst for image generation and editing
• Gemini Omni Flash 1.1 for video with audio
• Topaz Astra for video upscaling
• GPT 6 Astra as a model option for the Agent ($$$)
• An improved, simpler Model Selector
• New shot fields for Camera Format, Aperture and Framing
See you next week!
2
5,819
MiniMax H3 Max is live on @FilmSparkAI!
H3 Max and H3 Max Ref are now available for video generation at 480p or 768p with native audio. It generates fast (!!!) and it is priced below the rest of the video lineup, which makes it a real option for volume work and for iterating on a shot before committing to a final pass.
ALSO ... The Shot Panel has been completely REVAMPED!
This came out of user feedback. The goal was a panel that is easier to use and more compact, without losing anything the FilmSpark community already relies on.
The prompt now sits at the top for both the image and video stages, with everything that shapes it directly underneath.
Modify with Agent. Tap the button under the prompt, describe the change you want, and the agent rewrites the prompt, the fields, and the shot description in one pass. It uses the right prompting method for whichever models the shot is set to.
Adjust Shot. Change lighting, focal length, aperture, or art style from chips. Edits stack as pending changes so you can see what you asked for before it touches the prompt. Approve, and the prompt rewrites.
Hand editing. Type into the prompt whenever you want. Nothing gets blocked, and your text stays the source of truth for the rest of the panel.
Shot Description and Shot Direction are locked and synced, so they read back what the shot actually is.
Full walkthrough of the new panel in the video.
SEEDANCE PRICE DROP on FilmSpark.ai !!!!
We negotiated a new agreement with our inference partner and we're passing the savings directly to our fantastic users!
Every Seedance model is included: Seedance 2.0, 2.0 Fast, 2.0 Mini, and Seedance 2.5.
The new rates in practice: a 5-second Seedance 2.0 shot at 720p drops from 185 credits to 95. At 1080p, from 410 to 205. Seedance 2.5 at 720p goes from 57 credits per second to 29.
Seedance is the top-ranked video model on Artificial Analysis. Starting today, every FilmSpark plan gets roughly twice as many Seedance generations from the same credits. Nothing to change on your end. The new rates are already live.
#AIVideo #Filmmaking #Seedance #FilmSpark
Happy @FilmSparkAI Friday!
This Week: Seedance Upres Skill!
This week's demo is simple nature footage, and the point is how clean it looks at 1080p. Every shot started as a low res draft. Once the take was right, the shot was generated again at 1080p with the approved draft attached as a video reference. Seedance reads that reference for camera and timing, so the finished shot inherits the take you already approved instead of producing a new one. The main image carries the lighting and grade.
No upscaling. It's a second generation at a higher tier with the draft governing the motion, which is why the detail holds up.
Also This Week:
•Default video resolution and other advanced video model options can now be set when creating a project and in project settings.
•Import Shots, which brings shots from another project into a new or existing scene with the option to include shot details and referenced assets.
•Plus a lot more!
Bruce Lee vs Chuck Norris, restaged with Mozart and Bach.
The new Video Depth Anything model in @FilmSparkAI turns any video into a depth map you can use as a motion reference with Seedance 2.0. Live now in the Video Edit UI.
FilmSpark Agent has MAD SKILLZ ⚡
This week @FilmSparkAI released Skills: loadable playbooks that give the Agent real production expertise. Scene breakdowns. Storyboard boards. Character and location reference sheets. Prompt writers tuned for Seedance 2.0, Kling 3.0, and GPT Image 2. Use ours, or load your OWN custom Skills.
Give the Agent a scene and it assembles the shots into an annotated storyboard panel, then writes the multi-shot Seedance prompt to bring it to life. Script to storyboard to video, in one conversation.
The Agent runs on your own key: Claude, ChatGPT, or Gemini.
Watch the Agent + Storyboard Skill in action:
Made with @FilmSparkAI
@RocketLab's speed to mission is admirable, and we are here to serve with rad-tol and rad-hard MCUs operating from below the earth's surface to deep space. Congratulations, and our team is excited for all your next moves!
2
4,573
What we shipped at @FilmSparkAI this week:
→ Generate one shot across multiple video models, compare side by side
→ LTX 2.3 Quality models (incl. LoRA, Audio-to-Video, Reference-to-Video)
→ Ideogram 4 for images
→ Control Light to edit shadows and highlights
→ Per-shot sound effects (Mirelo SFX)
→ Reverse Video
We ship every Friday.
What we shipped at @FilmSparkAI this week:
→ Beeble Switch-X compositing: swap backgrounds, relight, change costumes, keep your subject intact
→ AI Agent can now read and edit your script directly
→ Credit estimates before every generation
→ Grok Imagine Video 1.5
→ Lyria 3 + Lyria 3 Pro music models
→ GPT 5.5 and Claude Opus 4.8
→ HiDream image models
We ship every Friday.
2
1
1,324
1/
Our biggest takeaways from @aionthelot this week:
Studios are done with general-purpose AI platforms. They want real production workflows.
Here's what that actually means.
2/
What the pros are asking for:
•Humans in charge of the creative
•Team workspaces for collaboration
•Enterprise security underneath
•Agentic workflows for real production
•Timeline and story-based production, not just nodes and canvases
3/
That last one is the missing piece in this market.
Almost no one is building around the story itself.
That's the part FilmSpark was built around from day one.
4/
FilmSpark is that workflow, available today.
We ship every week. We build from your feedback.
We're building this for you.
🤖 Made with AI
1
2
780
1/ The FilmSpark AI Agent is LIVE 🧵
In this video, we asked it to make a 30-second perfume commercial from scratch.
A full multi-shot commercial with consistent characters, scene structure, sequenced shots.
The Agent built it end to end.
2/ This is what an agent can do when it has a foundation underneath it.
Our agent operates on a structured production console.
3/ The Agent reads your story, generates assets, sequences shots, edits in place, resolves dependencies, writes its own prompts. Works across stories of any length.
Bring your own LLM — Claude, ChatGPT, or Gemini. Your API key.
4/ Live for all Pro and Ultimate subscribers right now.
4
901
The FilmSpark Agent is LIVE.
You often see agents making things from scratch, but there is so much more. In this video, we take one scene and make three changes across it:
1. Add a purple top hat to the detective in every shot
2. Change the wall color across all the shots in the scene
3. Swap the detective with a brand new female detective using GPT Image 2
The Agent shows you its plan first. You Accept or Reject. Nothing touches your project until you say go.
This is what an agent can do when it has a foundation underneath it. Our Agent reads your full story — script, scenes, characters, shot list, storyboard — and can analyze it, write prompts, and execute global changes across stories of any length.
And it works with the LLM you want , Claude, ChatGPT, or Gemini. You plug in your own API key.
This clip is just one scene with three changes. The Agent does much more: missing assets get generated automatically, dependencies resolve in the correct order, edits happen in place instead of regenerating, and complex multi-step operations sequence themselves.
The difference between this and "prompt in, video out" agents is the entire production console it operates on.
1/ A lot of platforms are shipping AI video Agents right now. Most do the same thing: prompt in, video out.
That's cool. It's also just the starting line.
Generation is the easy part. Here's the hard part 🧵
2/ What happens when you're 15 shots in and the client says "change the jacket to red"? What happens when you need to swap the lead across 30 shots? What happens when a shot needs a character that doesn't exist yet?
That's what the FilmSpark Co-Pilot Agent handles.
3/ It edits existing shots in place — "change the lighting to dusk" — without regenerating.
It resolves dependencies automatically — creates missing assets, then uses them.
It propagates changes across every shot that references an updated character. In order.
4/ Swap an actor across your entire project. One prompt. Bulk scene edits. Multi-step operations sequenced automatically. Full project awareness.
5/ And you see every step before it runs. Full execution plan. Approve or adjust. If one step fails, retry just that step.
Production isn't a straight line from prompt to output. It's iteration. That's what our agent handles.
6/ We're beta testing now. Want early access? DM us.
Also shipped:
• Massively improved cinematic shot prompts
• Actor turnaround + expression sheets
• Scene sidebar
• Seedance 2.0 at 15 sec
• XML export resolution + framerate options
1/ Friday ship from @FilmSparkAI 🧵
Multi-model image generation is here. Generate images with multiple AI models at the same time, then compare them side by side. Pick the best result per shot.
2/ This is what per-shot model selection looks like in practice. You're not guessing which model works best — you see them all and choose.
3/ Also shipped:
• 21:9 ultrawide aspect ratio
• Veo 3.1 Extend
• Team Workspace model controls for admins
• Multiple voice lines at once + voice line history
• Asset images as video references
• Collapsible left panel
4/ The Team Workspace model controls are a governance feature. If your team should only be using certain models, admins can now enforce that. Enterprise teams asked for this.
1
3
1,228
1/ Two weeks of shipping from @FilmSparkAI 🧵
#HappyHorse video models are here — strong new I2V and reference video generation. Check the sample.
2/ Also shipped:
• Kling 4K across three models
• GPT Image 2 (Low/Med/High)
• Video generation in Asset Studio
• HeyGen Lipsync Speed + Precision
• Scene-by-scene script segmentation
• Massive Library update with filtering + multi-upload
• Compare View for image versions
3/ The script segmentation is a bigger deal than it sounds. Your script is now broken into scenes automatically — review and refine each one before generating. Better structure in = better output out.
4/ Monthly credit contest is running. Best in Show: 2,500 credits. DM us or reply to our newsletter.
2
8
1,763