Shot 1 · 5s“The Reveal. Begin almost black. A narrow band of light travels across the marble, revealing the watch's silhouette and polished bezel. Slowly push toward the watch…”
☊ Some moments deserve your full attention.
2.7.5 adds twenty-six models and tools, and most of them make an existing job cheaper. Hand a clip a lighting reference and it is relit, with the subjects, motion, camera and sound exactly as shot. Swap the people in it for your own. Carry on past its last frame instead of restarting from a freeze. Nano Banana 2.1 undercuts the model it supersedes at every resolution, and video now starts at three cents a second. Free, like every update.

Until now a clip was finished when it came back. 2.7.5 adds passes that take the clip as the starting point: relight it, recast it, or keep going past the end of it. The rest of the release is catalog and price: a cheaper flagship image model, video from three cents a second, four new finishing passes in the Gallery and eighty more voices. Same release on Mac, Linux, Windows, in AssetLab Resolve and the Premiere beta.
Give H3 Max Relight your footage and a lighting reference, a sphere render or an HDRI-style preview, and it relights the shot. The subjects, the motion, the camera move and the sound all come back exactly as they were; only the light changes. From $0.08 a second at 768P. Apply stops before it spends if you have not attached a reference.
H3 Max Recast swaps the faces in a clip for your own, one photo per new person, in every shot the clip contains. The motion, the cuts and the audio stay put. It is the most expensive pass in the app at $0.30 to $0.45 a second, and the price is on the button before you commit.
H3 Max Extend continues a shot with new footage that keeps its motion and its sound, instead of restarting from a frozen last frame. The slider reads Footage to add, because that is what you are buying. $0.08 a second, or $0.04 on the Turbo tier.
Google's newest image model is cheaper than Nano Banana 2 at every tier: 1K from 4¢ against 8¢, 2K from 6¢ against 12¢, 4K from 13¢ against 16¢. It renders to 4K across fourteen aspect ratios, including the 4:1, 1:4, 8:1 and 1:8 banner crops nothing else here offers, and its editor takes up to fourteen references. Seedream 5.0 Flash joins it at 2.7¢ an image, six renders a run.
Grok Imagine 1.5 Lite is the cheapest video in the app at 3¢ a second at 720p, and unlike the full 1.5 it takes a text prompt as well as a start frame. H3 Max Turbo is the same H3 Max at half the rate. Also new: Vidu Q4 (a still animated with native audio, up to 4K), Seedance 2.5 US for data-residency work, and five fixed H3 Max looks with sound: VHS, Retro Toon 70s, Low Poly, Hand Drawn and 16-bit Pixel.
Reframe to… recomposes a still for a reel, portrait, square or wide crop instead of cutting it down. Split into layers returns the subject, props, surface and background as separate transparent layers. Clean up speech strips room noise off a voice recording, the first audio tool in the Gallery. And Kandinsky 6.0 upscales video without flicker between frames, usually cheaper than the route it sits beside.
ElevenLabs v4 Turbo reads the same twenty voices as v4 at half the rate, and Gemini 3.8 Flash and Flash Lite both come in under the Gemini voice you have now. That is 272 voices in all, still free to preview. Lyria 3.5, DeepMind's newest music model, is one flat fee a track. And H3 Max Lip Sync drives a face from your own audio at up to 2K, the highest any avatar model here reaches.
The version check now answers from the release manifest first, so a build always learns what the current version actually is. If you are on 2.7.0 through 2.7.4 and the button told you that you were up to date, it was wrong, and this is the release that fixes it.
Getting a shot right is half the job. 2.7.4 is the other half: captions written and embedded with the cut, a masked edit you can keep, a pack that lands where you want it, and a backup that brings the whole sequence back. Same release on Mac, Linux, Windows, in AssetLab Resolve and the Premiere beta.
Rendering a Director cut writes an .srt beside it and embeds the captions in the file, timed to the cut's real clock so a dissolve does not push them late. A shot with nothing spoken gets no caption, and pause cues never reach the screen. A masked edit now saves like any other render, a delivery pack lands in the folder you pick, and a backup carries your Director sequences with it.
Cut, copy, paste and select all in every text field, and copy or save any image. A two-finger tap on a trackpad is a right-click. Gallery cards get their own menu too: open, favourite, save a copy, reveal in Finder, delete.
Save a character, location or brand kit under a name already in use and it asks once: Replace it clears the old entry and its character sheet, or Keep both saves the new one as "Diner 2". Renaming into a taken name takes the next free one and tells you which it used, so your Cast stays readable.
Anything you save out of the gallery, one item or a whole selection, now opens a folder chooser instead of deciding for you. Your media vault is still the default, so nothing changes unless you want it to.
2.7.2 is about the moment after a render lands. Edit becomes a workspace big enough to look at, a mask keeps the rest of the frame exactly as it was, every voice honours a pause, and every button quotes its price before you press it.
The image you are editing fills the main area the moment you attach it, with scroll to zoom and drag to pan. When the result arrives, one button swaps between the original and the edit at the same zoom and the same position, so the only thing that appears to move is the thing that actually changed.
Paint over the part of the picture that is allowed to change. The model still edits the whole frame, and Asset Lab puts its work back only inside your stroke, with a soft edge. Everything outside is your original's own pixels, not a close match. Best for one garment, an object or a sign. Images only.
Write (pause) or (pause 1.5) anywhere in a line. Voices that understand a pause are told about it; voices that do not have the silence built for them, by reading the line in parts and joining them. Same cost either way, and the panel says which is happening. Write OCHRE [say: OCREEH] and the voice says the respelling while captions show the real word.
Rendering a cut writes an .srt beside it and embeds the captions in the file itself, timed to the cut's real clock so a dissolve does not push them late. Pause cues never show on screen, and a shot with nothing spoken gets no caption.
Sending a character, location or brand kit from a Blueprint now carries its reference images, not just its description, and tells the model the reference is the subject. The cast chips are on the Review step too, so a prompt drafted in one pass can still be cast before you send it.
Not just the prompt: the references, the frames, the guidance and the settings. Make it again repeats it exactly, and change one thing moves a single value with everything else held still, so the comparison actually means something.
All 82 models now quote from a local rate card instead of waiting on a usage estimate that a newly added model does not have. Ideogram V4.5 and the three GPT Image editors gained a Quality row, so a render that quietly cost 22¢ can cost 1.5¢ if that is what you want. Prices under a tenth of a cent read as "< $0.001" rather than "$0.00".
Choose where a pack is written instead of hunting for it afterwards. Select a Director sequence and the pack holds the finished cut, every approved take behind it, and a manifest carrying the full recipe and cast of each file.
Characters, locations and brand kits were tucked behind a tab of the prompt builder, the one workspace that barely uses them. 2.7 gives them their own: Cast, in the sidebar, with a count of everything in it. And what comes out of a sheet now looks like what went in.
Save a character from an illustration, an animation frame or a clay render and its sheet comes back in that medium. Every card has an Art style box, written for you the first time you open the sheet panel by reading the medium off your own reference, and yours to edit. It holds across all twelve views, into new scenes cast from the card, and into Director storyboards.
The printed board you pin up, for about 19¢. A character gets a turnaround on one baseline, six expressions, three poses, close studies of hair, collar and hand, and a colour palette. A location gets an establishing view, a numbered site plan with a key, approach, interior and landmark panels, a materials strip, and the place at dawn, blue hour and night. A brand kit gets a hero shot, product views, packaging and surface, hex values, and the product in context.
The Blueprint's first step now offers Character, Location and Brand kit together, each with a count, each filling the block it answers: a character the Subject, a location the Setting, a kit the Finish. And A location joins A person and A product as a subject in its own right, for an establishing shot or an empty set, with its own questions.
Check for updates and Support sit in the top bar. The check says plainly when you already have the latest, and nothing runs until you press it: no check on launch, no polling in the background. When a newer version has shipped it takes you straight to your Gumroad library, where the download is waiting. Updates stay free for life, as they always have.
Video is the expensive part. So 2.6 puts a step in front of it: Director plans the scene, draws a still for every shot, and lays the whole thing out as a board you can read at a glance, for cents a frame. Fix the shots that are wrong, approve the ones that are right, and only then buy the seconds.
Open The story in Director, choose the characters and the location you've saved, write a line about what happens, pick how many shots, and Plan shots writes the shot list: framing, angle, the action, a spoken line and who says it, and what each shot has to hand to the next. About 2¢ a plan, and every field is yours to edit before anything renders.

Each shot gets a Scene Still Frame drawn from the characters actually in it, the scene's first approved still and the frame before it, so faces, wardrobe, the set and the colour hold across the board instead of drifting. About 12¢ a still at 2K; Generate All Still Frames does the whole board in order.

▤ View Storyboard shows the approved stills two ways. The contact sheet is frames and labels, for judging the cut. The production board puts description, framing, location and duration under every frame: the version you send to someone else. Save a page as a PNG, or the whole board as a PDF.

An approved still isn't a sketch. It's the opening frame of the shot. → Shot by shot hands every approved frame to the video model you choose, so the take starts from exactly what you signed off. ⚡ Whole board → Seedance sends the entire sheet as one reference and returns a single multi-shot video (≈ 30¢ a second at 720p), a fast draft that interprets the board rather than copying it frame for frame.
≈ 2¢ a plan, ≈ 12¢ a still at 2K. A six-shot board is about 75¢ and a sixteen-shot board about $2. That's a fraction of one video take, and the number is on the button.
Choosing a character for a blueprint now opens the same thumbnail picker, and lists every character you've saved instead of the first six.
AssetLab Resolve 1.5 and AssetLab Premiere 1.2 ship the whole storyboard stage. With Auto-drop on, only the exported cut lands on the timeline.
Make it Talk opens from either side and its prompt buttons stay put. Launching Asset Lab twice now brings your window forward instead of starting a second copy with no key or library.
Same models, same prices, same license. What changed is where things live and how far your hand has to travel: the rail holds the workspaces in pipeline order, the render takes the middle of the screen, and the inspector on the right never scrolls the Generate button out of reach.
Workspaces sit in a rail on the left, in the order you use them: Prompt Lab, Image, Video, Audio, Edit, Director, Gallery. Your fal balance and this month's spend live at its foot. Every create workspace puts the result on the left and the controls in a right-hand inspector that scrolls on its own, so the Generate button and its price are always in view. The title bar stays put while you scroll.

Sequence setup (the look, the world, the sound) is decided once, so it now sits in the left column. The shots and the cut run down the right, with Preview and Render at the bottom. The two columns scroll independently, so moving through a long shot list never drags your setup out of view.

Open a render and everything you can do with it sits beside it, grouped by what it does: Analyze & fix, Edit, Animate, Extend, use as a reference, build a shot set; then Save as Character, Save prompt and Add to Brand Kit; then the Tools. No menu to open first. Favorite, Download, Show in Finder and Delete stay in the bar above.

Uniform cards with the prompt and cost under every render, instead of text that only showed on hover. Filters, projects, search and Select sit in one row above. "Favorites" is the one word for starred items everywhere.

One prompt gets you one beautiful, four-second clip. A film is shots in an order, with a voice and a score and a way from one moment into the next. Director is the room for that: you stop writing a prompt and start writing a shot list, and the app handles the rest of the assembly.
Model, aspect, resolution and the world every shot shares.
Up to six cards, each with its own prompt, length and price.
Generate takes, keep them all, click the winner.
One narrator, a line per shot, a soundtrack under it.
Level the voice and music, pick the transitions.
One finished file in your Gallery and on disk.
Every shot is a card with its own prompt and its own length. Generate a take and it stays on the card; generate another and that stays too. Click the one you want and it's the one that makes the cut. The rest wait, one click away, if you change your mind.

The top of every sequence is the part every shot shares. Pick the model once and every shot uses it. That's what keeps the grade, the lens and the texture from drifting between cuts. Change the resolution and every shot's price re-quotes on the spot.

Pick a narrator once (any of the 272 voices, or your own clone) and write a line on any shot. Hit Read and it's voiced for a fraction of a cent, placed exactly at that scene in the final mix.
(pause), or (pause 1.5) for a timed beat, anywhere in a line
Two sliders, ♪ Music and 👤 Voice, set the balance. You hear it live in the preview, and it's baked into the export exactly as you heard it. Between any two shots, pick how one hands off to the next. Then Export cut.


Saved sequences become 🎬 cards in the Gallery (filter: Sequences) with their takes, narration and mix intact. Click one to reopen it and keep editing. They travel with Back up / Restore.
This is the real sequence behind the film at the top of the page: the settings, the shot prompts and the narration exactly as they sit in Director, lightly trimmed for length. Nothing was edited outside the app.
Shot 1 · 5s“The Reveal. Begin almost black. A narrow band of light travels across the marble, revealing the watch's silhouette and polished bezel. Slowly push toward the watch…”
☊ Some moments deserve your full attention.
Shot 2 · 4s“Extreme close-up of the blue dial. Glide diagonally across the silver markers as a soft reflection reveals the sunburst texture. Shift focus onto the hands…”
☊ Look closer. Every detail draws you in.
Shot 3 · 5s“Low, three-quarter angle. Track slowly along the bracelet toward the case and crown. Cool highlights trace the brushed links and polished edges…”
☊ A quiet presence. A lasting impression.
Shot 4 · 6s“Return to the full watch on marble. Gently pull back, leaving space above it. At 0:17, fade in OCHRE, followed by OWN THE MOMENT…”
☊ Ochre. (pause) Own the moment. (comes in at 2s)
The four Seedance takes came to about $9 at fal's raw rates, the same numbers the Generate buttons showed before each click. The reference still was 16¢, the music bed 4¢, and every narration read, retakes included, under 3¢. The export, fades and all, rendered on the computer for free.
The narrator kept mispronouncing the brand, so the last line was typed the way it should sound,
OCREEH (pause) Own the moment., with a (pause) for the beat before the
tagline. Reads cost a fraction of a cent, so trying three spellings costs nothing.
A narrated walkthrough of a three-shot sneaker spot: open a sequence, choose the model once, write the shots, generate and approve takes, lay a track underneath, preview, export.
With sound, 2:23, recorded in Asset Lab 2.5. Also on YouTube.
Under every render in the lightbox, the last group of buttons is Tools: finishing passes that work on the clip or still in front of you instead of generating a new one. Pick one and a confirm bar opens with its options and the exact price for that item, measured from its real pixels and running time rather than a generic rate. Nothing is billed until you press Run.
Director's export is now two buttons. Render to Gallery keeps the finished film in your library, the way it always has. Render to Disk asks where to put it and writes exactly one copy there, for the cut that belongs in a client folder rather than your gallery.
Build a kit once in Prompt Lab → Library: up to eight anchor images plus a palette and your house rules. Apply it as a chip on the Image or Video tab and its notes join every prompt while its anchors ride along as references. Or cast it into a Director sequence.

Open any render in the lightbox and choose Save as Location… beside Save as Character. In Prompt Lab → Library → Locations, Build location sheet renders the place from seven angles (wide, reverse, left, right, high, low and a close detail) at ≈ $0.12 a view (≈ $0.84 for the set), with night, golden hour or your own view as extras.
Casting now sends the two images that matter, the main image and the stitched sheet, so a character and a location fit side by side in one render's references instead of one crowding out the other.
On the Kling 3.0 and O3 line, switch from Single prompt to Multi-shot and write up to six shots of 1–15 seconds each (3–15 seconds in total). The model renders them as one continuous take, cutting between your scenes.

An edit, an extension, an upscale: each render now knows what it was made from, and the lightbox shows the whole family as a strip, original first. Jump between versions without hunting the Gallery.

Select renders, choose Pack…, name it, and Asset Lab writes a folder of numbered files with a
manifest.json recording the prompt, model and seed of every asset. Hand it to a client, an
editor, or your future self.

Filter models by what you're making: Product accuracy, Readable text, Character continuity, Cheap drafts; Cinematic, Native audio, Start → end frames for video.
(pause) and (pause 1.5) work in the Audio tab too, with exact silence on MiniMax (every clone included) and ElevenLabs.
A pre-ship pass re-verified every Generate button against fal's published rates and corrected the ones that had drifted. Details below.
The Resolve editions ship all of it. With Auto-drop on, only the exported Director cut lands on your timeline.
The price on the button is the promise, so when fal's rates move, the buttons move with them. Here's what 2.4 corrected, and what it fixed.
| Model | What changed in 2.4 |
|---|---|
| Seedance 2.5 References | Now prices attached reference clips: fal bills them at 0.6× the rate over the clips' length plus the output, so a 5 s shot at 720p with 30 s of reference clips is ≈ $9.93, not the $2.37 shown before. |
| Seedance 2.5 at 1080p | $1.164 per second; fal now publishes the rate. |
| FLUX.2 [flash] | Adds $0.005 per attached reference image. |
| Grok Imagine images | $0.02 (was quoted $0.022, which is the edit rate). |
| Sonilo music | Starts at 10 s, since fal bills anything shorter as 10 s anyway. |
| AssetLab Resolve tools | Trellis 2 3D models are $0.30 (were quoted $0.05), Hunyuan 3D 3.1 Pro shows its price ($0.525 with PBR materials), and Create LUT is $0.12 (was quoted $0.08). |
The shot plan is about 2¢. Each still frame is about 12¢ at 2K, so a six-shot board runs roughly 75¢ and a sixteen-shot board about $2, before you've bought a single second of video. Every price sits on its button before you press it.
No. A brief on its own is enough, and the planner will invent the cast and the place. Saved characters and locations are what make the same faces, wardrobe and set hold from panel to panel, and the Details you write on them beat the reference photo on hair, clothing and accessories. The face always comes from the photo.
Two ways. → Shot by shot makes each approved still the opening frame of its shot, on whichever video model you choose. ⚡ Whole board → Seedance sends the whole sheet as one reference and returns a single multi-shot video (≈ 30¢ a second at 720p). It's a fast draft that interprets the board rather than copying it frame for frame. Either way the takes bill at the model's raw rate; see the rate card, and stills vs video for when to storyboard at all.
Included. Director and its storyboards ship in Asset Lab 2.7.5 on Mac and Linux and in AssetLab Resolve 1.6.5, at no extra cost. Updates are free for life: download the new version from your Gumroad library, and your existing license key unlocks it.
The takes bill at the video model's raw fal.ai rate, shown on every Generate button before you click (see the rate card). Narration reads cost about $0.002–$0.01 a line. Exporting is free when the cut has transitions or fades (it renders on your own computer) and a few cents for a plain-cut sequence stitched on fal. The 20-second OCHRE spot above cost about $9 in takes.
38 of the 40 video models in the catalog: every model that generates from a prompt, including Kling 3.0, Seedance 2.5, Veo 3.1, FLUX 3, Wan 3.0 and LTX 2.5. You choose one per sequence so the look holds. Every model that takes reference images (Seedance 2.5 References, Gemini Omni References 1.1 and the reference versions of Seedance 2.0, Kling O3, Veo 3.1, Wan, MiniMax H3, Grok and HappyHorse) carries them into every shot, and 27 models can open each shot from an anchor frame or chain from the previous shot's last frame.
Up to sixteen shots. Each shot's length follows what the chosen model accepts (Kling 3.0 takes 3–15 seconds, Seedance 2.5 up to 30), and the running total flags anything over 30 seconds as long for a first cut.
No. Different tools, similar name. The Director panel on the Video tab turns cinematography (genre, palette, lighting, camera move, lens) into one-click direction for a single shot, and it's still there. The Director tab arrived in 2.4: it assembles many shots into one finished cut.
Yes. AssetLab Resolve 1.5 ships the same Director tab, storyboards included. With Auto-drop on, only the exported cut drops onto your timeline; takes, still frames and narration reads stay in the Gallery, where → Timeline sends any of them on demand. More on the Resolve page.
The Edit canvas, storyboards, Cast, Director and the whole 2.7.2 studio: $49 once, on every Mac and Linux machine you own.