Making video with CSuite.
Every setting, media row, AI action and editing tool in the video workspace, with a figure for each, followed by recipes that chain them into finished shots. The short version lives on the main guide; this page is for when you want to know exactly what a control does.
Get around
The workspace
Open Video from the rail. This workspace is about single clips: generating one, continuing it, and cutting it. For stitching clips, audio and titles into a longer piece, see Compositions.
- 1 · Sidebar
- Your project folder pill, the + that clears the selection, and every MP4, WebM, MOV and AVI in the folder as a frame with its length, newest first. A “Generating…” row sits at the top while a clip is being made.
- 2 · AI menu
- Two extend actions, a render step for a draft clip, two upscales and two background removals, run on the clip you have open.
- 3 · Editing toolbar
- Trim, Resize, Crop, Speed, Audio, GIF and Format, each running on your machine through the bundled ffmpeg.
- 4 · Info and Model
- Info lists duration, resolution, frame rate, codecs and bitrate plus the model and prompt behind a generated clip. The Model button opens the picker and that model’s settings.
- 5 · Preview
- The clip, fitted. With a clip selected on a model that takes a source video, it is also what Extend works from.
- 6 · Transport
- Play, a scrub bar and the time, then Loop, a preview-only playback rate and the camera button that saves the current frame.
- 7 · Composer
- The prompt, the wand, a row per kind of media the selected model takes (start and end frame, references, source video, audio), the active settings and cost estimate, and Create.
The + beside the folder pill (“Clear selection”) deselects the clip so the next prompt creates a new one; it is disabled with “Select a project folder first” until a folder is set. Esc clears the selection too, as does clicking Video in the rail again.
Files
- Thumbnails are a frame from each clip, made on your machine, with the length overlaid. A tile that cannot be read shows a film glyph instead.
- More actions on a row: Show in Finder (Show in File Explorer on Windows, Show in file manager on Linux), Duplicate and Delete, which asks first. Delete or Backspace with a row selected also deletes.
- Add your own footage by dragging MP4, WEBM, MOV or AVI files onto the list (“Drop video to add it here”). They are copied into the project folder and get the same tools and AI actions as generated clips.
- Filter and sort appear above 20 files: “Filter videos by name…” and a sort button cycling Newest, Oldest and Name.
- Names: a generated clip takes the first two words of its prompt, title-cased; an edit saved as a new copy adds a suffix that says what was done (
-trim-1-6,-speed-1.5x,-mute,-remove-bg); a rendered draft adds its resolution; a clash gets “ 2”, “ 3”.
Generate
Models and settings
The Model button at the top right opens the picker. Cloud models run through Runware, Replicate and OpenRouter on your platform keys, or direct on a vendor key (Google for Veo, Runway, Kling, ByteDance for Seedance, MiniMax for H3, xAI, Black Forest Labs). Open-weight models run on the bundled GGML runtime. Every model with its samples, prices and limits is in the catalog. Video is billed per second of output and is the most expensive modality, so the cost hint beside Create matters here.
- Aspect ratio: the shapes the model accepts (16:9 and 9:16 on most; 21:9, 4:3, 1:1 and 3:4 on Seedance and H3). A model that edits an attached clip follows the clip’s shape.
- Duration: a fixed ladder (4, 6 or 8 s on Veo; 5 to 12 s on H3) or a free range (4 to 30 s on Seedance 2.5). Longer is proportionally dearer.
- Resolution: 480p, 720p, 1080p, on some models 2K or 4K. A 1080p second costs several times a 480p one on the same model.
- Audio: On or Off on models that can render a soundtrack (Veo 3.1, Seedance 2.5, Kling 3.0, LTX-2.5, Flux 3 Video, Wan 3.0); P-Video-2 Pro always does and shows no switch. On changes the price on several models.
- FPS (LTX, P-Video-2, Seedance 1 Lite), Camera (LTX-2.5’s named moves: static, dolly, jib) and Fixed camera (Seedance 1 Lite) appear only where the model has them.
- Draft: a cheap low-resolution tier on Seedance 2.5, Seedance 1.5 Pro, Flux 3 Video and P-Video-2; see Drafts.
- More settings: a seed, a negative prompt, a quality mode and whatever else the model takes.
Under Auto the picker alone shows and the settings ride in the prompt: “8 seconds, with audio, 9:16” sets length, sound and shape for that run and never reaches the model. Auto prefers Runware, then Replicate, then OpenRouter, or a downloaded local model with no key, and anything you attach narrows the pick to a model that can take it. Values are remembered per model.
Generating
A failed run keeps its error above the box with Retry and your prompt in place. Every generated clip remembers its model, prompt and settings; the Info popover shows them and can refill the composer.
Start and end frames
Models that animate an image show a Start frame row in the composer; attach a photo or a generated still and describe the motion. Many also take an End frame, generating the transition between your first and last shot. Each becomes a thumbnail chip with a remove badge, and both clear when you change model, since the slots differ. Picked images are staged beside your project; a project PNG (a captured frame, for instance) is used as it is.
Reference media
Beyond the two frames, the composer grows a row for each kind of media the selected model takes, each capped at the model’s published limit and each explaining itself on hover:
- Reference: images that guide characters, objects or style (up to 30 on Seedance 2.5, seven on Gemini Omni Flash, nine on H3).
- Source video: the clip the model edits or extends (Veo 3.1 on Runware, Seedance 2.5, Gemini Omni Flash, Flux 3 Video and the two Flux Video models).
- Ref. video: clips that guide motion and camera (Seedance, H3, Wan 3.0, Kling 3.0 on Runware).
- Ref. audio and Audio track: a guide for voice or rhythm, or a track the video syncs to (Seedance, H3, LTX-2.5, P-Video-2).
A row a source clip rules out greys itself out with the reason (“This model can’t take frames alongside a source clip”), and the other way round. On a model that takes both a source clip and reference videos, one picker serves both: the first file is the source, the rest are guides. Under Auto every row is offered with neutral labels (“Video”, “Audio”), and whatever you attach becomes a hard requirement on the model Auto picks. Attached chips show the file itself; a video chip shows its poster frame.
The wand
The wand at the start of the composer’s toolbar rewrites a one-line idea into a proper shot: subject and action, camera movement, framing, lighting, setting and mood, pinned to a single continuous shot. It runs on your text model and puts the rewrite back in the box. Since a clip costs real money and minutes to come back, a sharper prompt first is cheaper than a second attempt.
It reads the composer. With a start frame attached it writes motion direction instead of re-describing what the frame already shows; with an end frame it adds the transition. Sound is written only when the model renders audio and the switch is on, with dialogue in quotes where the vendor documents it, and forbidden otherwise. With several references attached it cites them as [Image 1], [Image 2]. On a model with its own negative prompt field it keeps exclusions out of the main prompt.
Drafts
Some models render a cheap low-resolution draft first. With Seedance 2.5 or Seedance 1.5 Pro on Runware, or Flux 3 Video or P-Video-2 selected, switch Draft on in the popover. The Resolution locks to the draft tier (480p on Seedance, 720p on Flux 3) with the reason shown, keeping your stored choice for when Draft is off. A draft costs around a quarter of the full price.
When a Seedance or Flux 3 draft is the one you want, open the clip’s AI menu and choose Render in 1080p (Render in 720p on Seedance 1.5 Pro): the same shot is rendered again at full resolution on the provider that made the draft and saved beside it with the resolution in its name. Prompt, length and audio carry over; nothing is re-prompted. P-Video-2 and Flux 3 on a vendor key offer the Draft switch without the second step. The render action also appears on a draft placed in a composition.
Local generation
Wan 2.1, Wan 2.2, LTX-2.3 Distilled and H3 Local run on your own machine through the bundled GGML runtime (built on stable-diffusion.cpp) on Apple silicon, Windows x64 and Linux x64. Download under AI model providers; weights can be paused, resumed or cancelled, shared parts download once, and each model lists the RAM and disk it needs.
- Controls: Steps (8, 20 or 30), Resolution where the model has more than one size, Duration (capped at 2 s on Wan 2.2 TI2V-5B), an aspect ratio, a seed and a negative prompt. Wan 2.2 TI2V-5B, LTX-2.3 and H3 Local take a start frame.
- Time: video is the heaviest local workload. Wan 2.1 1.3B takes about 23 minutes for a 2 s clip on Apple silicon; the 14B Wan is hours per clip and listed for high-end CUDA hosts. The status line names the phase and the sidebar shows the run.
- Memory: the runtime offloads to the CPU and tiles the decode when the weights or the frame count are large for the GPU; that is slower, not broken. A decode that ran out of memory is reported as a failure, never saved as a grey clip.
- Output is written as MP4 through the bundled ffmpeg, so local clips get every tool and AI action below.
Continue and transform
Extending a clip
Two items lead the AI menu above a clip:
- Extend (use video) attaches the whole clip as the composer’s source video (“Clip attached”) for a model that continues footage. It appears only when the selected model takes a source clip, or under Auto, which then picks one that does.
- Extend (use last frame) grabs the clip’s final frame and seeds it as the start frame (“Start frame set”), so it works with any model that animates an image. Pick it on a model that takes no start frame and a message asks you to switch.
Both stop at loading the composer: you write what happens next and press Create. The new footage lands beside the original as its own clip. On Runware a model returns the original plus the continuation, and CSuite trims it back to the new footage (“Trimming the extension…”), because the new part is what chains: extend again and the continuation is the next source. Google caps a Veo chain at 30 seconds in total; the long cut is assembled in a composition, where the same two actions exist on a video clip.
Upscale and remove background
Below the divider, four jobs that need no prompt: Upscale to 1080p and Upscale to 4K (Real-ESRGAN), and Remove background (green screen) and Remove background (transparent webm) (BiRefNet, which handles people, products and animals). The result is saved as a new file beside the original and becomes the selection.
- They run on Replicate only: the selected provider when that is Replicate, else your first connected Replicate key. With none, the action shows “Video upscale and background removal run on Replicate. Connect it under Models to use them.” with an Open Models button.
- Green screen gives an MP4 with the subject on solid green, which every editor can key. Transparent webm keeps real alpha and composites correctly over other layers in a composition, but QuickTime, Safari and most editing suites flatten it, which is why both exist.
- One action at a time per clip; while one runs the menu leads with Stop, and a second choice is refused with a message. Nothing in the menu is ever disabled.
Edit and upscale models
Two models in the picker work on a clip rather than a prompt. Flux Video Edit (Runware) takes a Source video and an instruction (“make it night, keep the camera move”) and returns the edited clip at the same length. Flux Video Upscale (Runware, Replicate) takes a source clip and an Upscale factor from 1.5× to 3×, with a creativity dial on Replicate; its prompt is optional, so Create enables as soon as a clip is attached. Neither takes a ratio, size or duration, since the output follows the clip, and both refuse to run with no clip attached. Both are priced by the clip’s length, and the upscaler by its pixels too.
Play and edit
Player, frames and info
- Transport: play, a scrub bar and the time. Space plays and pauses; ← and → step the playhead a second, Shift with an arrow steps one frame. All of it is ignored while your focus is in a text field.
- Loop and a preview-only playback rate (0.5×, 1×, 2×) sit after the time and stay set as you move between clips. Neither changes the file.
- Capture frame: the camera button saves the frame under the playhead as a PNG beside the clip (“Saved … — find it in the Image workspace”), a quick still or a start frame you can attach by hand.
- Info (the i) lists Duration, Resolution with its tier (“1920 × 1080 px (1080p)”), Frame rate, Video codec, Audio (or “none”), Bitrate, Size, Format and Modified, plus Model and Prompt for clips CSuite made. Use prompt & settings refills the composer: the prompt always, the settings when the same model is still selected.
Every tool below opens an options row under the header with a Save a new copy switch and Apply, runs one at a time through the bundled ffmpeg with its own busy label and percentage (“Trimming (42%)”), and never leaves your machine, however the clip was made. On, the result is a sibling with a suffix; off, the file is rewritten in place and the player reloads it. Ten-bit or high-chroma sources are written back as standard 8-bit so they play everywhere, and HDR sources are tone-mapped.
Trim
Type a Start and End in seconds, or park the playhead and press Set beside either field. The row reads “Keeps 5.5s” and the kept range shows as a band on the scrubber; an impossible range reads “Enter a valid range”. Apply trim cuts frame-accurately (the clip is re-encoded, not stream-copied, so the cut lands exactly where you set it). A copy is named with the range, such as -trim-1-6.
Resize and crop
- Resize: pick a height of 240, 360, 480, 720 or 1080 and the width follows the clip’s aspect ratio. It never enlarges; use an upscale for that.
- Crop: pick a Ratio (Free, or 1:1, 4:3, 16:9, 9:16 and the rest of the app’s list, the same one the image crop uses) and drag the marquee over the preview; drag inside it to move it. Apply crop writes the result; a marquee below the minimum is refused (“Crop area is too small.”). The tool waits for the clip to load before it enables.
Speed and audio
- Speed: 0.5×, 0.75×, 1.25×, 1.5× or 2×. The soundtrack is re-timed with its pitch preserved, unlike the audio workspace; the range stops at 2× for that reason. A copy is named
-speed-1.5x. - Audio → Extract MP3 pulls the soundtrack into a new MP3 beside the clip (“find it in the Audio workspace”), ready for the audio tools or transcription. A clip with no audio track says so.
- Audio → Remove audio strips the track without re-encoding the picture, so it is instant and lossless; a copy is named
-mute.
GIF and format
- GIF: choose 10, 12, 15 or 20 frames per second and a width of 320, 480 or 640 pixels, then Export GIF. It always makes a new file (“Saves a new GIF to Images”), with a palette computed from the clip so colours hold up.
- Format: convert the container between MP4, WebM and MOV. The current one reads “Already MP4” and is refused (“File is already in that format.”); turn on Save a new copy for a same-format copy. WebM is the one that can carry transparency.
Recipes
Animate a still
Draft cheap, render once
Sound on a silent clip
A cutout for a composition
Local first, cloud last
Reference
Keyboard shortcuts
⌘ is Ctrl on Windows and Linux. The main guide lists every workspace’s shortcuts together.
Player
Space- Play or pause the selected clip
← / →- Step the playhead one second
Shift+← / →- Step one frame
Esc- Clear the selection (outside a text field)
Files and composer
Enter / Shift+Enter- Create / add a line to the prompt
Delete / Backspace- Delete the selected clip (asks first)
⌘+ / ⌘- / ⌘0- Zoom the interface in, out, or back to 100%
Troubleshooting
- “Connect a provider under AI model providers to generate videos.” Add Runware, Replicate or OpenRouter, a vendor key, or download a local model.
- “None of your connected providers include video models in this workspace.” Your keys are for platforms without video; add one that has it.
- “Video upscale and background removal run on Replicate.” The transforms need a Replicate key whatever model you generate with. Press Open Models.
- “An AI action is already running on this clip.” Wait, or press Stop at the top of the AI menu.
- “The selected model doesn’t take a start frame.” Extend (use last frame) needs an image-to-video model; pick one that shows a Start frame row.
- Extend (use video) is missing. The selected model takes no source clip. Pick one that does (Veo 3.1 on Runware, Seedance 2.5, Gemini Omni Flash) or Auto.
- “Extending is available in the Video workspace.” You are in a composition; use the clip’s actions there or come back here.
- A media row is greyed out. Something you attached rules it out on this model; hover for the reason, or remove the other attachment.
- Create stays disabled on Flux Video Edit or Upscale. They need a source clip attached; a prompt alone does nothing.
- The bar sits at 95%. It is an estimate on Runware and OpenRouter, not a measurement; the run is still going. Only Replicate, a Runway key and local downloads report a real percentage.
- The extension came back shorter than the clip. That is correct: on Runware the result is trimmed to the new footage, which chains as the next source. Assemble the full cut in a composition.
- Render in 1080p is missing on my draft. It appears only on a Seedance or Flux 3 draft made on Runware; P-Video-2 and vendor-key drafts have no finish step.
- The transparent webm looks flat in my editor. Most editing suites and QuickTime drop VP9 alpha. Use the green-screen variant there and key it.
- “Failed to extract audio.” The clip has no audio track; Info shows “none”.
- “Crop area is too small.” Drag a larger marquee before Apply crop.
- “File is already in that format.” Pick another container, or turn on Save a new copy.
- “Only video files (MP4, WEBM, MOV, AVI) can be added here.” Convert other formats first.
- The + beside the folder pill is disabled. “Select a project folder first”: pick a folder from the pill.
- A local model is missing on this machine. The GGML runtime has builds for Apple silicon, Windows x64 (CUDA) and Linux x64 (Vulkan) only.
- A local clip takes very long. Expected: tens of minutes for a 2-second clip on Apple silicon, hours for the 14B model. Lower the steps or the resolution, or draft locally and render in the cloud.
One clip, every feed