Install SlydeForge

✨ New — Assistant request queue

• No more waiting for the AI: while a request is running, "Send" becomes "Add to queue". Your requests go out one at a time, in order, as soon as the AI is free. "Slide 3" still means the slide you were looking at when you typed.
• No more waiting for your review either: results wait for your approval in a non-blocking combined review window, approved in order (D1, D2…), action by action if needed.
• On the thumbnails: D1, D2… badges on modified slides, added slides shown dotted, and an orange "Conflict" badge if you edit by hand a slide that a pending request also changes. Read-only Preview of the presentation with all pending changes.
• Validate automatically (option): each result is applied as soon as it is ready, with one undo step per request.

✨ New — Changing a specific area of a slide

• On a busy slide (dashboard, several tables or blocks), a request only changes the targeted area; the rest of the slide stays as it is.
• When in doubt, the assistant asks: the possible areas are numbered in orange on the slide, or you draw it yourself ("I'll draw it on the slide").

✨ New — Live dictation

• The text appears while you speak, like on a phone keyboard, in every prompt box. No Internet needed, in French, English, Spanish, German and Italian (about 40 MB per language, installed once).
• Spoken punctuation ("comma", "period", "new line"…) and automatic capitals. An EN ▾ button switches language in one click.
• "Prepare the microphone" on the first click on "Dictate": checks that the microphone is selected and set up.

✨ Improved — Microphone sound

• A close, clear voice in recordings, where it used to sound "far away": right channel picked, light noise reduction, less room echo, standard loudness for spoken videos.
• Guided microphone setup in 15 seconds (Settings › Voice and dictation… › Microphone), with a clear verdict and before/after listening. Microphone meter in the presenter console while recording.

✨ New — SVG images

• SVG images accepted everywhere (insert, drag and drop, brand kit, brief, assistant), including a drawing pasted from Inkscape or Figma.
• Change the color of an SVG icon: Properties › Image › Drawing color, brand kit or custom color. The assistant can do it too ("make the icons accent-colored").

🧭 Changed — Settings menu

• Organized in 4 sections: Artificial intelligence, Content and media, Application, Updates. New Royalty-free photos…, Voice and dictation… and Language and appearance… windows (with dark mode).
• Clearer AI… window: "Advanced" section, "Assistant" tab, no more cut-off fields.

🐛 Fixes

• "Add a green background to slide 3" gave a red background: the requested color is now respected.
• "Turn slide 3 into a table" no longer produces an empty slide: automatic retry, otherwise the slide stays unchanged with a clear message.
• No more false "figures missing from the sources" alert on 01, 02, 03… numbering.

📖 User guide

• Guide (app and website, French and English) updated: request queue, specific area, live dictation, reorganized Settings, new screenshots.

✨ New — Copy / paste

• Copy slides from one project to another, including between two SlydeForge windows: Ctrl+C on the thumbnails, Ctrl+V in the other project. Notes, animations and transitions come along; a single Ctrl+Z undoes the paste. Slides keep the look of their source project, with an "Adapt to project theme" banner one click away.
• With PowerPoint, both ways, plus Office-style paste options (keep source formatting, use project theme, picture). Drag thumbnails between windows to copy.
• Paste content from elsewhere: images (screenshots, browser), formatted text (Word, mail, web, PDF), PowerPoint shapes, Excel or web tables (a real, editable table), a full Word document with its images and tables. Drag an image or text straight onto the slide.
• "Duplicate" (Ctrl+D) no longer overwrites the clipboard.

✨ New — Routines (formerly "Agents")

• "Agents" are now "Routines" (Routines menu, /routine; /agent still works). Your existing routines are kept.
• {{Name}} variables: the information is requested each time you use the routine (text, date, number, list, web address, files), with remembered values and per-client value sets.
• ✨ Improve: the AI challenges the routine's prompt. ✨ Translate also translates the variables. A "Client presentation" routine is provided.

✨ New — Brief and sources

• 🔍 Challenge the brief: the AI restates what it understood, proposes an outline and asks at most 4 questions before generating.
• File › Extract project brief: rebuilds a standalone brief that reproduces the same presentation from a project you have edited.
• Blocked web pages (LinkedIn…): assisted reading — open the page inside SlydeForge, frame the useful info and pick up the photo.
• Addresses written "www.…" without https:// are now read along with your request.

✨ New — AI usage

• "AI usage" window (Settings, Subscription menu, /consommation): all consumption by period, project and AI, with plain-language explanations, a detailed log, CSV export, a personal monthly budget and non-blocking alerts.

🐛 Fixes

• "Rewrite slide 7 as a table and write its notes": the rewrite is no longer lost; the notes describe the new content, follow your tone and duration, and outdated notes are flagged.

✨ New — Spell checker

• The spell checker no longer uses AI by default: in the brief, the chat, agents…, it now relies on the Windows spell checker — free, offline, instant, with no tokens used. Nothing changes in use: mistakes underlined in red, one-click correction, "Ignore". It also catches repeated words ("the the") and suggests the word closest to what you typed.
• Advanced AI correction, optional (Settings › AI › "Advanced AI spell checking with this engine", unchecked by default): it also catches grammar, agreement and homophone errors, but uses tokens at every typing pause. The choice is remembered for each AI engine — for example on for a local model, off for an engine that bills its tokens.

🐛 Fixes — Generating from a brief

• Much faster generation start with a local model (Ollama): spell checks on the brief could pile up and run before the presentation outline, causing waits of up to several minutes. Only one check now runs at a time, and it stops as soon as you click "Generate".
• The brief dialog closes immediately when you click "Generate" (it could freeze for about 2 seconds).
• Waiting explained: if the AI model hasn't started answering after 20 seconds, the assistant explains why (local model still loading, or busy with a previous request).

ℹ️ Good to know

• Updating from within SlydeForge: nothing to do, as usual.

🐛 Fixes

• Component installation: frozen progress bar and "SlydeForge is not responding": after updating from an older version, leftover launcher settings slowed the component installation down considerably and froze the startup window on "Preparing download…". This is fixed: a full installation takes about 40 seconds again.
• More informative progress bar: it now moves steadily through every step — looking for components, downloading file by file (name and size shown), then installing, the longest step.

ℹ️ Good to know

• Updating from within SlydeForge: nothing to do, as usual.

🎨 AI Theme Studio — the highlight of this release

• Create a theme with AI (Theme menu, ✨ Create with AI… button in the theme editor, ✨ New theme… button in the brief): only a short text is required (business, mood). Optional: website, colors (enforced or as inspiration), logo, images, icons, existing documents (PPTX, PDF, screenshots), expected layouts, fonts and format.
• 3 clearly different proposals in a single AI call, shown as real slides with content drawn from your business. Click to choose, double-click to see every layout, Ctrl+click to blend two proposals (0 tokens) or ✨ Refine with a free-form request.
• Cost shown before you start, then the amount actually used.
• Your website read in under 20 s: colors actually displayed, fonts, logo, photos, business. A site that blocks automated reading is reported, never bypassed.
• Step-by-step progress, with a Stop button.
• Consistent quality whatever the AI model: palettes with computed contrast, 8 graphic styles (banner, rules, diagonal, soft shapes, frame, dot pattern, offset blocks, full-page photo), checked layouts (overflow, broken words, logo legibility).
• Fonts: a missing font is replaced by a close one, and you're told; a Google Font installs in one click, without admin rights.
• Logos legible everywhere: several variants per theme (color, white, monochrome…), each slide automatically picks the most legible one for its background, on screen and in PowerPoint.
• The created theme applies its style to all generated slides.

✨ New

• AVIF images accepted everywhere (theme, image insertion, brief, chat, workflows).
• Thumbnails and drag-and-drop for files in the Studio, and a large view of each proposed layout.

🚀 Startup and installation

• New, lighter SlydeForge.exe launcher: SlydeForge starts directly, with no command window or hidden step. In Task Manager, the app now shows up as "SlydeForge".
• Fewer antivirus alerts: the new startup avoids a sequence that some antivirus products (including Kaspersky) wrongly took for malware. The launcher now carries its publisher and version information.
• Faster startup: components checked in ~0.3 s instead of about 4 s, roughly halving the time between the click and the loading screen.
• SlydeForge-branded startup window when components need installing (first launch), with a real progress bar and a clear message if something goes wrong.
• First launch explained: if Windows shows "Windows protected your PC", SlydeForge explains once why, and how to stop seeing it. A "LISEZ-MOI - Premier lancement.txt" file (French and English) is included in the .zip archive.

🐛 Fixes

• Updates: no more command window or full reinstall of components (~310 MB) after each update; an interrupted update is offered again on next launch.
• Much faster first installation: 34 s instead of almost 3 minutes.
• "Failed to load Python DLL" error when changing language.
• A second SlydeForge window wouldn't open if one was already open.
• Small windows flickering at startup; invisible close button on the loading screen.
• Some AI providers (Copilot GHE) were wrongly treated as unable to see images.
• Studio: a reference page (Wikipedia, LinkedIn…) no longer gives its own logo to the theme; an error after the AI response no longer loses the paid call ("Rebuild (0 tokens)").
• Team layout: member roles now meet the minimum readability contrast.

ℹ️ Good to know

• Updating from within SlydeForge: nothing to do, as usual.
• Installing from the .zip archive: before extracting it, right-click the .zip file › Properties › tick Unblock › OK. Windows then won't show a warning at launch.

🎬 Motion engine — the highlight of this release

• The slideshow really plays animations: fade, rise, pop, bounce, wipe, mask reveal, exits, pulse… Text stays sharp, a click during an animation moves to the next step, and going back or jumping to a slide shows it without replaying anything.
• Morph transition: between two slides, shared elements glide, resize, change color and rotate instead of disappearing. Exported as a real PowerPoint Morph (with a fade fallback for PowerPoint 2016, LibreOffice and Google Slides). Duplicate for Morph from a right-click.
• Kinetic typography and living data: text revealed line by line, word by word or letter by letter, a counter on a figure ("12,500 customers", "98.6%"…), highlight, charts and timelines that draw themselves, a slow Ken Burns zoom on the background image.
• 10 motion styles (Corporate sober, Keynote, Tech neon, Editorial luxury, Cinema, Data storytelling…) in 3 clearly distinct intensities (Subtle, Elegant, Spectacular), and 4 animated backgrounds in your brand-kit colors.
• Motion window (Ctrl+Shift+M): looping animated preview of the slide, a plain summary of what the style does, applied to the whole deck, the current slide or the selected slides. ▶ Play button in the editor to replay a slide without starting the slideshow.
• "Animate slide (AI)" knows all these effects, with a before/after preview and lines you can uncheck; the chat understands "put a counter on the figure" or "do a Morph with the previous slide".
• Faithful PowerPoint export: native effects, "with / after previous" triggers finally respected, fully editable file. "100% PowerPoint" mode so the slideshow and the file play exactly the same thing.
• Check motion: non-blocking warnings (animation too long, slide empty until the click, too many strong effects…).

✨ New

• Group / ungroup objects (Ctrl+G / Ctrl+Shift+G): move and resize together, kept as real groups in PowerPoint.
• Edit several objects at once from the properties panel (text, colors, outline, shadow, animation, position…), without touching anything else on each object.
• Better-looking tables: style gallery in your brand-kit colors (Grid, Clean, Banded, Accent, Minimal), solid or underlined header, total row, density, vertical alignment, readability fixed automatically.
• Slide language in the brief: language selector (5 languages) saved in the project, then respected by AI edits and speaker notes.
• GitHub Copilot Enterprise: list of the models your organization actually allows, a clear message when a model is refused, and "Test connection" tests the configured model.
• Simpler thumbnail context menu: 8 entries grouped by purpose instead of 17, with no feature removed.

🔒 Subscription

• From the Essential tier: personal AI connectors (OpenAI-compatible API, GitHub Models, Copilot Enterprise), motion styles and animated backgrounds. Classic animations and transitions, Morph and "Simple animations" stay free; an already staged deck plays and exports normally.
• Free trials (Gemini, Groq, OpenRouter) capped at 300,000 tokens per month without a subscription, uncapped with one.

🎯 Better results with local models (Ollama, gemma4…)

• No brand-kit slide delivered empty anymore: a poorly filled slide is retried on its own, otherwise its content goes into an equivalent layout.
• Response format enforced for local models: tables, steps and comparisons finally filled with the real data.
• The model's context window is detected and respected; "Test connection" explains how to enlarge it, based on your actual setup.
• Thinking mode no longer makes generation fail: a step that runs out of room is redone without thinking.

🐛 Fixes

• Reliable token count: the amount shown is exactly the amount counted, reasoning and images included; the cost of a cancelled action stays visible.
• PPTX export: no more white table rows with unreadable text.
• Lists shown in brackets (["…", "…"]) on some slides.
• "Custom…" color in the properties panel never applied.
• Dark theme: disabled items are now greyed out.

🎯 New generation engine — the highlight of this release

• The AI writes, SlydeForge lays out: the AI now only writes the content; the layout is computed from your brand kit (grid, typography, measured contrast, no overflowing text). Measured on an 8-slide presentation: 13.6 s instead of 152.7 s, about 17 times fewer tokens, no layout defect detected (24 before).
• Your brand kit's slide templates guide the result in every format (16:9, square, portrait) and are preferred by the plan.
• Visible progress: the plan's titles, then each slide, appear as they arrive, at no extra cost.
• A complete result from the very first generation (checkboxes in "New from a brief"): transitions and animations (free, no AI call) and speaker notes.
• Your brief is really used: attached images are placed on slides (portraits on a "team" slide), web links cited are actually read as sources, and royalty-free photos are kept far more often.

✨ New

• Two-step Formatting: other structures, then bold art directions (value-based color coding, title band, depth, status icons…) as real previews. ✨ Refine bar for a free request or to combine two proposals, 🔍 Enlarge to compare full size, Create a new slide checkbox to keep the original.
• Challenge asks questions first: a diagnosis, then up to 3 one-click questions before rewriting the slide; about 70% fewer tokens, and figures or dates lost or added are flagged.
• Photo button (/suggest_image): up to 6 royalty-free photos already built into the slide, with a readable credit.
• AI-generated images: AI image button (/generate_image) and a "Generate AI images" checkbox in "New from a brief", with an upfront cost warning (requires a configured image generation service).
• Search the whole presentation (Ctrl+F): slides, tables, charts, timelines and notes, with highlighting, a counter and a list of matches.
• Local model check (Ollama…): "Test connection" explains in plain words whether the model is suitable (speed, thinking, vision) and how to pick another one.
• Zoom in the before/after preview: click an image to compare full size.

🐛 Fixes

• Generation stuck for minutes with a local "reasoning" model (gemma4, qwen3…): thinking is now turned off automatically (8-slide presentation in 19.8 s instead of 53.9 s).
• Spell check: the chosen correction replaces exactly the misspelled word, and clicking in the text works normally again.
• "Team" slide left empty despite attached photos; extra empty circle in a brand-kit agenda.
• Only one art direction offered on a slide based on a brand-kit template.
• A Challenge rewrite can no longer make the slide poorer (lost photo, icons or cards) without warning you.
• Numbers better centered in badges; table and chart text more faithful on screen; bullets properly spaced in the PowerPoint export.

✨ New

• Tables: per-axis borders and custom sizing — row and column separators can now be shown independently (horizontal only, vertical only, both, or neither), and multiple selected columns or rows can be resized at once to a precise width or height.
• Hyperlinks now exported to PowerPoint: text carrying a link stays clickable during the presentation, not just in the SlydeForge preview.

🐛 Fixes

• Multi-series charts turning monochrome, or unreadable text on a dark background: each series gets its own distinct color again (preview and PowerPoint export), and pie chart label text now automatically switches between white and black based on the background color.
• Double-clicking an element (timeline, table, chart…) sometimes ALSO opened the slide background properties — fixed.
• The custom color picker would sometimes reopen itself right after confirming a choice — fixed.
• Buying a paid marketplace component: now redirects straight to the slydeforge.com pricing page, without asking for an email address inside the app.