What changed in each version, newest first. The same notes are in the app under Settings → What’s new.
Latest version 0.8.2
October 5, 2026
Windows and Mac
Lotzy 0.8.2
Latest
Security (Windows): your API keys are now stored encrypted with Windows data protection in Lotzy’s own data folder. They used to sit as plain text in Documents/Lotzy, which OneDrive may sync. Existing keys move over by themselves the first time you open this version; nothing to re-enter. (Versions before 0.8.2 will not see the moved keys.) On a Mac, keys stay where they were for now.
Security: the Lotzy window is locked to Lotzy’s own page and only runs Lotzy’s own code (content security policy); it asks for no camera, microphone or location, and a dropped file or link can no longer take the window elsewhere.
Security: the local AI engine download is checked against its expected checksum before it is unpacked; a changed file is deleted.
Security: the offline period can no longer be stretched by setting the computer’s date back, and an installed Lotzy does not start with debugging switches.
Security: the AI connection (MCP) compares its key in constant time and only answers under its own local address; LoRA links are accepted from Civitai and Hugging Face only.
Lotzy 0.8.1
Local AI: more open models to download and run on your own computer. Video: Hunyuan Video 1.5 (text or image to video, 480p) and LTX-2.3, which makes the sound with the picture. Image (new for Local): Z-Image Turbo (very fast), FLUX.2 klein 4B (also edits with up to 4 references), Qwen Image 2512 and Qwen Image Edit 2511. Settings → Local AI lists them in Video and Image groups, with a fit score and time for your machine.
LoRA: a new Settings → LoRA page. Paste a Civitai or Hugging Face link (or drop a .safetensors file) and Lotzy reads which model the LoRA was made for, finds both halves of a Wan 2.2 high/low pair, shows its trigger words and the Lotzy models it fits. In the composer the LoRA button sits in the model settings with a strength for each. Local models download a LoRA once; fal.ai models (the new LoRA versions of Wan 2.2, Z-Image, FLUX.2 klein, Qwen Image and FLUX in Settings → Models) get it as a link. A LoRA that turns out not to fit a model is marked and no longer offered for it.
Local AI: Wan 2.2 14B Lightning runs again. Its speed-up LoRAs stored a value the engine could not read and the engine stopped at once (code 3221226505); Lotzy now converts them once after the download, and also files you copied in yourself.
Local AI: images with ş, ı, ü or other non-English letters in their name (or in the Windows user folder) no longer stop the engine (“code 1”). When a local run fails, the card now shows the engine’s real reason and the full log is saved to last-error.log.
Local AI: a Steps setting next to Resolution. Fewer steps are faster, more steps add detail; the time estimate follows. Wan 2.2 14B starts at 4 (split between its high- and low-noise halves), Turbo 5B at 8, 5B at 20.
Local AI: Wan 2.2 14B runs in 4 steps (2 high noise + 2 low noise), as its Lightning authors set it: about 40% faster, with closer time estimates.
Chat: a step the assistant corrected on its own (a wrong name, a model it looked up again) is no longer shown in red as if something failed; it reads “tried another way”, with the details on hover.
New Terms of Use (version 2026-10-05), asked to be accepted again: Lotzy’s role as a tool, the licenses of models and LoRAs (e.g. Hunyuan outside the EU, UK and South Korea; LTX-2 above USD 10M revenue), a fuller list of what is not allowed, disclosing AI content where the law requires it, and reports of misuse.
Lotzy 0.8.0
Local AI: make videos on your own computer, free of credits. Settings → Local AI looks at your graphics card, scores each open model (Wan 2.2 Turbo 5B, Wan 2.2 5B, Wan 2.2 14B Lightning) for it and estimates how long a 5 s video takes. Press Download and Lotzy downloads and sets up everything by itself in the background, no ComfyUI or setup. The models then appear under “Local” in the composer and work from the prompt box and the chat; generating with a model that is not downloaded yet downloads it first and starts by itself.
The source switch is now API · Account · Local: API holds your Higgsfield and fal.ai models together, grouped in the menu.
Chat is smarter and cheaper. It knows every model before it writes for it: what the reference images do (some models take none), limits, billing and how to prompt it. It sees what you have selected on the canvas, so “animate this” just works. It keeps project notes (say “remember that Isabella’s hair is short”) and a summary of long conversations. Auto sends simple requests (move, rename, tidy, list) to the provider’s small model. Pictures are looked at in small size, many at once on one contact sheet. Ready-made buttons: shot list from an idea, 3 variations, animate the selection, put the videos in order, tidy the canvas.
Chat: next to Prompt there is now Chat. The prompt box makes room and a chat opens on the right of the canvas: talk to Claude, GPT or your custom model (with your API keys) and it works in Lotzy for you, with the same abilities as the AI connection. Every project keeps its own conversation, and generations are approved on a card in the chat with the credits of each shot and the total.
Chat: point at things on the canvas. In Chat mode the @ button on an item, “Add to chat” or dragging it onto the chat attaches it to your message, and typing @ in the message picks one by name, like in the prompt box. Images and videos both work; only their names are sent, and the assistant looks at them when it needs to.
Chat is far cheaper to run: short tool answers, earlier results trimmed, the provider’s prompt cache and low reasoning effort. The chat shows the tokens of the last answer and of the whole conversation.
fal.ai: open models such as Wan, LTX, Hunyuan, Flux and Qwen, usually much cheaper per clip. Add your fal.ai key in Settings → API keys; the models appear as a third source next to API and Account, and any image or video model on fal.ai can be added by searching in Settings → Models.
Settings → Models: choose which models the composer lists. The most used ones are on by default, so the menu stays short; switch any model on or off, per source, or go back to the defaults.
The Settings window has a side menu and keeps one size; only the page scrolls.
AI connection (MCP): Claude and other AI apps can now use Lotzy for you. In Settings → AI press Connect next to Claude Desktop and press Install in Claude’s window, no terminal, no restart (Claude Code, Cursor and VS Code connect from the same place). Then ask in plain words: look at the canvas, arrange it, write notes, move items between projects, generate images and videos, cut the edit, export MP4 or XML.
Your credits stay in your hands: before any generation the AI shows the list with the credits of each shot and the total, and asks you. A batch is approved once, not shot by shot; Lotzy also asks once more in its own window (you can turn that off).
It works only on this computer while Lotzy is open, protected by a key. Everything the AI does appears live on the canvas and in Edit, and Ctrl+Z undoes it.
Edit: Ctrl+F plays the preview full screen (Esc or Ctrl+F again to leave), also with the new button next to the transport.
Edit: the Track Select Forward tool, as in Premiere. Press A (or Select forward under the timeline) and click a clip: it and every clip after it on all tracks are selected and move together when you drag; hold Shift for one track only. V or Esc goes back to the selection tool. Both shortcuts can be changed in Settings → Shortcuts.
Lotzy 0.7.27
Edit: the sound of your videos now shows on its own lane under the video tracks, with real waveforms (audio files too). Videos without sound leave the lane empty.
Keyboard shortcuts you can change, in Settings → Shortcuts (or the Shortcuts button in Edit). In Edit: Z zooms in, X zooms out, R cuts every track at the playhead, Q / W cut the part before / after the playhead and close the gap, \ fits the whole timeline.
Edit: long videos fit. You can zoom out until the whole edit is on screen (Fit button or \), and the ruler shows hours.
Edit: drag videos, images and music straight from Explorer or Finder onto the timeline. They land where you drop them and join the project, so videos and images show on the canvas too. Settings → Files decides whether they are copied into the project or used from where they are; this now applies to music as well.
Edit: select several clips by dragging a box over empty space (Shift or Ctrl adds), then drag one of them to move them all. Drag with Alt to copy the clips on the timeline; the copies use the same source files, nothing is copied on disk. Dragging the playhead near an edge now scrolls the timeline.
Canvas: send images and videos to another project, from the right-click menu or “Send to…” in the selection bar. They arrive on that project’s canvas with their prompts and settings; linked files stay linked, and the originals stay where they are.
Canvas: right-click → “Show in the whole project” keeps an image or video on the canvas of the main project and every folder inside it, wherever you marked it (characters, locations, style frames). Nothing is copied; elsewhere it sits in a row at the top with a quiet grey frame and the name of its folder.
Canvas: Ctrl+C (⌘C on a Mac) copies the selected images and videos as files: paste them into a folder, a chat or another Lotzy project. Right-click → “Show in folder” opens the folder with the file selected.
Lotzy 0.7.26
Mac: “Check for updates” now finds new versions. It looks at the latest Lotzy release and “Download the new version” downloads the DMG for your Mac’s chip; open it and drag Lotzy onto the one in Applications. (Automatic install on the Mac needs a signed app, which comes with the public release.)
Lotzy 0.7.25
Settings → Files → “Add files without copying”: images and videos you drop on the canvas or add with “Add media” stay where they are and are shown from there. Removing one from the canvas never deletes the original; a small link mark shows which cards are linked.
Mac: two fingers move the canvas, pinch or ⌘+scroll zooms (there is no middle button); Space+drag still moves.
Mac: the system’s own close, minimise and full-screen buttons, full screen from the green button or ⌃⌘F, and a Mac menu so ⌘C / ⌘V / ⌘Q work.
Mac: new installs keep projects in ~/Lotzy.
Lotzy 0.7.24
Smoother feel: switching between Canvas and Edit is instant (the editor stays open in the background); windows and menus fade in, a pressed card lifts a little, new results grow in, and Arrange / Align move the cards to their places instead of jumping.
One fixed top bar for the whole window (logo, Canvas | Edit, scene, account and window buttons); only the area below it changes.
A small Lotzy loader while the app starts and the first time Edit opens, instead of a stutter.
Projects instead of scenes: add sub-folders, rename (F2 or double-click) and delete from the right-click menu; delete asks whether to only remove it from Lotzy or move the folder to the Recycle Bin. Drag a project onto the edge of another to reorder, onto its middle to put it inside, or onto the empty part of the list to bring it to the top level.
Prompt: “See all” opens a tall prompt box (Esc closes it); @ references can be picked with the arrow keys and Enter, and show as chips with their thumbnail in the text.
Save a video frame as an image: right-click a video on the canvas for its last frame, first frame or the point you paused it at, or press “Frame to canvas” in Lotzy Edit at the playhead; the full-size PNG lands next to the video.
Restoring an older generation finds its reference images again when they have moved (another save folder or project), and a missing image shows a clean placeholder.
Higgsfield account: Seedance 2.5 takes several reference images (omni-reference mode is switched on by itself).
Premiere / DaVinci XML: videos without a sound track no longer show as missing.
Animations use only the graphics card and stay short; with “reduce animations” turned on in Windows they are off.
Lotzy 0.7.23
Lotzy Edit (preview): a full-window editor: switch between Canvas and Edit at the top left. Multi-track timeline with trim, split, snapping, crossfades, titles and music; undo and autosave per scene.
Right-click items on the canvas → Send to Lotzy Edit. In the editor, right-click a clip to generate it again or to generate a continuation from its last frame.
Agents: the Editor cuts and orders your shots from a sentence; the Writer turns an idea into a shot list and fills missing shots by generating them.
Export an MP4 (rendered on your computer with hardware video decoding, with titles, crossfades and sound) or an XML timeline for Premiere Pro and DaVinci Resolve.
Seedance 2.5 draft: turn on “Draft 480p” for a quick take at about a quarter of the price; press “1080p” on its card when you like it. With the Higgsfield account the same take is finalized at 1080p (same prompt, references and seed), within 7 days.
Lotzy remembers the models you pick (and your prompt assistant); recently used models are at the top of the list.
Source switch: with both an API key and a Higgsfield account, choose API or Account next to the model menu to list only those models.
Smoother canvas with many items: cards redraw only when they change, and media outside the view is not loaded.
New installs save projects to C:\Lotzy instead of Documents (often synced by OneDrive). Existing folders stay where they are.
Lotzy 0.7.22
Connect your Higgsfield account: Settings → API keys → Connect Higgsfield account. Sign in once in your browser and Lotzy generates with your subscription credits, no API key needed. Higgsfield’s own model list appears under “Higgsfield account · credits”, the cost shows in credits, and your remaining credits are in the top bar.
An API key still works as before; with both, the model menu shows both groups.
Lotzy 0.7.21
Feedback with screenshots: attach up to 3 images, paste one with Ctrl+V or drag it in. Images are shrunk before sending.
Everyone who leaves an email with their feedback now gets a short thank-you from [email protected].
Lotzy 0.7.20
Custom provider for the prompt assistant: use OpenRouter, Groq, Together, DeepSeek or a local model (LM Studio, Ollama). Settings → API keys → Custom provider: fetch the model list, tick the models you want, or type a model name; they appear in the prompt assistant menu. Thanks for the suggestion!
Admin: fixed the oversized card for feedback sent from the app.
Lotzy 0.7.19
Arrange everything: one click puts the whole canvas in a neat grid, first generation to last, all the same width. “Previous layout” brings your own arrangement back any time.
Right-click menu: arrange, align (left, centre, right, top, bottom), space evenly, same size, group, notes, add to prompt and delete.
Groups: selected items move together inside a named frame (Ctrl+G, Ctrl+Shift+G to ungroup; double-click the name to rename).
Notes: sticky notes on the canvas for scene ideas, with colours.
Undo and redo for canvas changes (Ctrl+Z, Ctrl+Y).
Lotzy 0.7.18
Automatic updates: new versions download in the background and install when you close Lotzy.
Settings → General → Updates: check for updates yourself, see progress, and switch between automatic and manual updates.
New “What’s new” tab with the notes for every version.
Lotzy 0.7.17
Access can now be approved automatically when the Lotzy team turns on auto-approval; waiting accounts open by themselves.
Lotzy 0.7.16
Seedance 2.5 now supports 1080p on all three routes, with the correct 1080p price in the cost estimate.
Lotzy 0.7.15
Lotzy is now downloaded from lotzy.ai/download; updates come from the same public source.
Lotzy 0.7.14
Terms of Use and Privacy Policy: accepted once when you sign in and in the installer.
Lotzy 0.7.13
Feedback button in the top bar: send bugs and ideas to the team without leaving Lotzy.
Lotzy 0.7.12
Sign in with Google or email. New accounts join the alpha waitlist and Lotzy opens by itself once approved.
Lotzy 0.7.11
English interface by default, with a language setting (English / Türkçe).
Lotzy 0.7.10
The app is now called Lotzy.
Lotzy 0.7.9
Every current Claude model in the prompt assistant, including Opus 5.5 and Fable 5.1.
Lotzy 0.7.8
New premium red design.
Lotzy 0.7.5
Exact cost before every generation, and the price shown on every result on the canvas.
Alpha