Skip to content

Free, open-source AI video generator for Mac. Run LTX-2.5 and MiniMax H3 (Hailuo) locally on Apple Silicon: text-to-video and image-to-video with sound, lip-sync, character LoRA training, songs from lyrics (YuE2) and image editing (Qwen-Image-Edit). One-click install via Pinokio. Offline: no cloud, no API key, no NVIDIA GPU.

fix: settings save was a silent no-op; new endpoints sent multipart, server only parses urlencoded

mrbizarroc6c90add202d
diff --git a/launch/07_settings_loras_tokens_announcement.md b/launch/07_settings_loras_tokens_announcement.md
new file mode 100644
index 0000000..f9b96c1
--- /dev/null
+++ b/launch/07_settings_loras_tokens_announcement.md
@@ -0,0 +1,155 @@
+# Phosphene update announcement (settings + LoRAs + tokens)
+
+Three drafts. Voice: casual, direct, light comic timing, no marketing
+buzzwords, owns mistakes openly. **Pick one** and paste — or mix.
+
+Recommended: **Option B** (short thread). Hero is short enough to land
+on its own, and the replies give people who care the technical detail
+without bloating the timeline.
+
+---
+
+## Option A — single punchy tweet (~280 chars)
+
+```
+Phosphene update.
+
+Renders are 7× smaller by default — X uploads work now. LoRAs are in (HDR is one click, CivitAI browser built in). No more .env files for tokens, paste them in Settings.
+
+Update in Pinokio. Make weird shit. 🎬
+```
+
+256 chars. Good for casual X drop. Use when you don't want to thread.
+
+---
+
+## Option B — short thread (RECOMMENDED)
+
+### Hero (245 chars)
+
+```
+Phosphene update — shipped a lot this weekend.
+
+Renders are 7× smaller by default (X uploads work now), LoRAs are in (HDR is one click, CivitAI browser built in), and no more .env files anywhere.
+
+Update in Pinokio. 🎬
+
+🧵 details
+```
+
+### 1/ — the codec confession (276 chars)
+
+```
+On output: I was shipping yuv444p crf 0 lossless by default. Great for color grading, terrible for X — which straight up rejects yuv444p with a misleading error.
+
+That was a design mistake. Default is now Standard (yuv420p crf 18, ~7 MB per 5s clip). Lossless is still there.
+```
+
+### 2/ — what "still there" means (251 chars)
+
+```
+The lossless preset is now called "Video production" because that's who actually picks it — colorists, editors, VFX folks who'll grade or composite downstream.
+
+If you just want a clip that plays on X, Standard is your default and you don't have to think about it.
+```
+
+### 3/ — LoRAs (276 chars)
+
+```
+LoRAs work end-to-end now.
+
+— HDR: one-click toggle (it's the official Lightricks LoRA fused at load time, you don't need to know that)
+— Drop any .safetensors in mlx_models/loras/ → shows up in the picker with strength slider
+— Built-in CivitAI browser: search, preview, install
+```
+
+### 4/ — no env files (276 chars)
+
+```
+Asking normal users to set CIVITAI_API_KEY and HF_TOKEN in their shell felt wrong. Most people don't know what an env var is.
+
+Settings modal now has a tokens section. Paste your CivitAI key + HF token once, stored locally, never sent anywhere except the actual auth headers.
+```
+
+### 5/ — credits + close (236 chars)
+
+```
+Update via Pinokio Discover or hit Update if you've already got it. Click ⚙ for tokens + output presets, expand LoRAs in the form, browse CivitAI in-app.
+
+Big thanks to @cocktailpeanut for catching the gnarliest bugs along the way.
+
+Make weird shit and tag me.
+```
+
+---
+
+## Option C — Discord / Reddit / Pinokio post (long-form)
+
+```
+Phosphene update — shipped a lot this weekend, three things you'll feel:
+
+**1. Renders are 7× smaller by default.**
+The output was lossless yuv444p crf 0. Great for color grading, terrible
+for X — which straight-up rejects yuv444p with a misleading error. That
+was a design mistake on my end. Default's now Standard (yuv420p crf 18,
+~7 MB per 5s clip, plays everywhere).
+
+The lossless preset is still there, renamed "Video production" because
+that's who actually picks it. Plus a Web preset for smallest files and
+a Custom mode if you want a specific pix_fmt + CRF.
+
+**2. LoRAs work end-to-end.**
+- HDR is a one-click toggle next to the Enhance button. (It's actually
+ the official Lightricks HDR LoRA fused at pipeline-load time. You
+ don't need to know that.)
+- Drop any .safetensors into mlx_models/loras/ and it shows up in the
+ LoRA picker, with a strength slider per row.
+- Built-in CivitAI browser: search filtered to LTX 2.3 LoRAs, preview
+ the actual animations (most LTX LoRAs ship video previews), Install
+ button drops the file in mlx_models/loras/ with the right metadata
+ sidecar. Auto-enables the LoRA after install.
+
+**3. No more .env files.**
+Asking normal users to set CIVITAI_API_KEY and HF_TOKEN in their shell
+felt wrong. Most people don't know what an env var is.
+
+Settings modal now has a tokens section. Paste your CivitAI key + HF
+token (read access is enough), click Apply. Stored locally in
+panel_settings.json, never sent anywhere except the actual auth
+headers to civitai.com / huggingface.co.
+
+Power users can still override with shell env vars — settings win when
+both are set.
+
+---
+
+Update via Pinokio Discover or hit Update if you already have Phosphene
+installed. Click the ⚙ in the header for token + output settings,
+expand the LoRAs section in the form, browse CivitAI in-app.
+
+Big thanks to @cocktailpeanut who caught the gnarliest bugs along the
+way (silent helper crashes, audio regressions, the whole start.js URL
+thing).
+
+Make weird shit and tag me at @AIBizarrothe.
+```
+
+---
+
+## Voice notes for the next announcement
+
+What I'm matching here:
+
+- **Open with the user-felt outcome, not the implementation.** "Renders
+ are 7× smaller" lands harder than "we now ship yuv420p by default."
+- **Own mistakes openly.** "That was a design mistake on my end" reads
+ honest; "we've optimized the encoding pipeline" reads like marketing.
+- **One throwaway joke per beat, max.** "You don't need to know that"
+ about the HDR LoRA. "Make weird shit" as the close. Don't pile.
+- **No buzzwords.** No "powerful", "robust", "seamless", "reimagined".
+- **Concrete numbers.** "7×", "5s clip", "yuv420p crf 18" — anchors the
+ claim.
+- **Credit collaborators by handle, not by role.** @cocktailpeanut, not
+ "the Pinokio creator."
+- **Close with action, not signature.** "Update in Pinokio." or "Make
+ weird shit and tag me." Not "Cheers!" or "Looking forward to feedback."
diff --git a/mlx_ltx_panel.py b/mlx_ltx_panel.py
index 0b310f8..052c725 100644
--- a/mlx_ltx_panel.py
+++ b/mlx_ltx_panel.py
@@ -5763,7 +5763,12 @@ async function clearToken(which) {
if (which === 'civitai') fd.set('civitai_api_key', '');
if (which === 'hf') fd.set('hf_token', '');
try {
- const r = await fetch('/settings', { method: 'POST', body: fd });
+ // urlencoded body — see applySettings for why.
+ const r = await fetch('/settings', {
+ method: 'POST',
+ headers: {'Content-Type': 'application/x-www-form-urlencoded'},
+ body: new URLSearchParams(fd),
+ });
const data = await r.json();
if (!r.ok || data.error) {
alert('Could not clear: ' + (data.error || `HTTP ${r.status}`));
@@ -5817,7 +5822,16 @@ async function applySettings() {
const hfInput = document.getElementById('hfTokenInput').value.trim();
if (hfInput) fd.set('hf_token', hfInput);
try {
- const r = await fetch('/settings', { method: 'POST', body: fd });
+ // Convert FormData → URLSearchParams so the body is sent as
+ // x-www-form-urlencoded — the panel's parse_qs only understands
+ // that wire format, NOT the multipart/form-data fetch sends by
+ // default with FormData. This bug silently turned every settings
+ // save into a no-op (server saw empty payload) until caught.
+ const r = await fetch('/settings', {
+ method: 'POST',
+ headers: {'Content-Type': 'application/x-www-form-urlencoded'},
+ body: new URLSearchParams(fd),
+ });
const data = await r.json();
if (!r.ok || data.error) {
status.textContent = data.error || `HTTP ${r.status}`;
@@ -6027,7 +6041,11 @@ async function deleteLora(path, name) {
const fd = new FormData();
fd.set('path', path);
try {
- const r = await fetch('/loras/delete', { method: 'POST', body: fd });
+ const r = await fetch('/loras/delete', {
+ method: 'POST',
+ headers: {'Content-Type': 'application/x-www-form-urlencoded'},
+ body: new URLSearchParams(fd),
+ });
const data = await r.json();
if (!r.ok || !data.ok) {
alert('Delete failed: ' + (data.error || `HTTP ${r.status}`));
@@ -6192,7 +6210,11 @@ async function civitaiInstall(btn, item) {
fd.set('download_url', item.download_url);
fd.set('meta', JSON.stringify(item));
try {
- const r = await fetch('/civitai/download', { method: 'POST', body: fd });
+ const r = await fetch('/civitai/download', {
+ method: 'POST',
+ headers: {'Content-Type': 'application/x-www-form-urlencoded'},
+ body: new URLSearchParams(fd),
+ });
const data = await r.json();
if (!r.ok || !data.ok) {
const status = document.getElementById('civitaiStatus');