feat: add YouTube importer, copyright checks, and optimize LLM viral

.env.example

@@ -2,3 +2,4 @@ DEEPGRAM_API_KEY=your_deepgram_key_here
 ANTHROPIC_API_KEY=your_anthropic_key_here
 LLM_PROVIDER=deepseek
 DEEPSEEK_API_KEY=
+GROQ_API_KEY=

src-tauri/src/lib.rs

@@ -54,9 +54,11 @@ async fn environment_status(state: tauri::State<'_, AppState>) -> Result<Environ
         has_gemini_key: std::env::var("GEMINI_API_KEY").is_ok(),
         has_openai_key: std::env::var("OPENAI_API_KEY").is_ok(),
         has_openrouter_key: std::env::var("OPENROUTER_API_KEY").is_ok(),
+        has_groq_key: std::env::var("GROQ_API_KEY").is_ok(),
         llm_provider,
         has_local_whisper_model,
         has_ollama,
+        has_ytdlp: media::command_exists("yt-dlp"),
     })
 }
 
@@ -483,6 +485,14 @@ async fn generate_candidates(
                 .await
                 .map_err(to_command_error)?
         }
+        "groq" => {
+            let key = api_key
+                .or_else(|| std::env::var("GROQ_API_KEY").ok())
+                .ok_or_else(|| "Set GROQ_API_KEY or supply Groq API Key to generate candidates.".to_string())?;
+            llm::detect_candidates_with_groq(&normalized, &key)
+                .await
+                .map_err(to_command_error)?
+        }
         _ => {
             let key = api_key
                 .or_else(|| std::env::var("DEEPSEEK_API_KEY").ok())
@@ -678,12 +688,80 @@ pub fn run() {
             set_selected_clip_count,
             render_flat_clip_for_candidate,
             delete_project,
-            rename_project
+            rename_project,
+            check_youtube_copyright,
+            download_youtube_video
         ])
         .run(tauri::generate_context!())
         .expect("error while running AutoShorts");
 }
 
+#[derive(serde::Serialize)]
+pub struct CopyrightCheckResult {
+    is_safe: bool,
+    license: Option<String>,
+}
+
+#[tauri::command]
+async fn check_youtube_copyright(url: String) -> Result<CopyrightCheckResult, String> {
+    let output = std::process::Command::new("yt-dlp")
+        .args(&["--dump-json", &url])
+        .output()
+        .map_err(|e| format!("Failed to run yt-dlp: {}", e))?;
+
+    if !output.status.success() {
+        return Err("Failed to fetch video metadata from YouTube.".to_string());
+    }
+
+    let json_str = String::from_utf8_lossy(&output.stdout);
+    let parsed: serde_json::Value = serde_json::from_str(&json_str).map_err(|_| "Failed to parse yt-dlp output".to_string())?;
+
+    let license = parsed.get("license").and_then(|v| v.as_str()).map(|s| s.to_string());
+
+    let is_safe = if let Some(lic) = &license {
+        lic.to_lowercase().contains("creative commons") || lic.to_lowercase().contains("reuse allowed")
+    } else {
+        false
+    };
+
+    Ok(CopyrightCheckResult { is_safe, license })
+}
+
+#[tauri::command]
+async fn download_youtube_video(url: String) -> Result<String, String> {
+    let downloads_dir = dirs::download_dir()
+        .ok_or_else(|| "Could not find Downloads folder".to_string())?;
+
+    let output_template = downloads_dir.join("AutoShorts_%(id)s.%(ext)s");
+    let output_template_str = output_template.to_string_lossy().to_string();
+
+    let output = std::process::Command::new("yt-dlp")
+        .args(&[
+            "--format", "bestvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best",
+            "--merge-output-format", "mp4",
+            "-o", &output_template_str,
+            "--print", "after_move:filepath",
+            "--no-simulate",
+            &url
+        ])
+        .output()
+        .map_err(|e| format!("Failed to run yt-dlp: {}", e))?;
+
+    if !output.status.success() {
+        let stderr = String::from_utf8_lossy(&output.stderr);
+        return Err(format!("yt-dlp failed: {}", stderr));
+    }
+
+    let stdout = String::from_utf8_lossy(&output.stdout);
+    let filepath = stdout.lines().last().unwrap_or("").trim();
+
+    if filepath.is_empty() || !std::path::Path::new(filepath).exists() {
+        return Err("Could not locate downloaded file".to_string());
+    }
+
+    Ok(filepath.to_string())
+}
+
 fn project_dir(state: &AppState, project_id: &str) -> PathBuf {
     state.data_dir.join("projects").join(project_id)
 }
@@ -893,12 +971,9 @@ fn build_drawtext_filters(
         let mut font_option = String::new();
         for path in &font_paths {
             if std::path::Path::new(path).exists() {
-                let escaped_path = path
-                    .replace('\\', "\\\\")
-                    .replace(':', "\\:")
-                    .replace('\'', "\\'")
-                    .replace(' ', "\\ ");
-                font_option = format!("fontfile={}:", escaped_path);
+                // Windows FFmpeg requires a single backslash to escape the colon even inside single quotes
+                let clean_path = path.replace('\\', "/").replace(':', "\\:");
+                font_option = format!("fontfile='{}':", clean_path);
                 break;
             }
         }

src-tauri/src/llm.rs

@@ -36,12 +36,16 @@ pub async fn detect_candidates_with_deepseek(
 ) -> Result<Vec<CandidateDraft>> {
     let segments = compact_segments(&transcript.segments);
     let prompt = format!(
-        "You are identifying the most viral moments and strongest short-form clip candidates from a long-form transcript. \
-For each candidate, the clip must be self-contained, starting with an extremely engaging hook within the first 3 seconds (to capture immediate attention on social feeds), \
-30-90 seconds long, and cut at clean sentence/thought boundaries. Favor highly shareable content: concrete stories, \
-strong opinions, emotional turns, surprising or counter-intuitive claims, clear payoffs, and high-energy/dramatic peaks. \
-Avoid rambling setup, context-dependent references, and pure filler. Return up to 10 candidates as JSON matching this schema: \
-{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}\n\nTranscript:\n{segments}"
+        "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection, but extract AS MANY highly viral moments as possible. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
+Clips should be 30-90 seconds long, completely self-contained, cut at clean boundaries, and deliver a massive payoff (a mind-blowing fact, hilarious joke, highly controversial opinion, or deep emotional insight). \
+Return up to 25 candidates as JSON matching exactly this schema: \
+{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}
+
+Transcript:
+{segments}"
     );
 
     let response = reqwest::Client::new()
@@ -111,12 +115,16 @@ pub async fn detect_candidates_with_gemini(
 ) -> Result<Vec<CandidateDraft>> {
     let segments = compact_segments(&transcript.segments);
     let prompt = format!(
-        "You are identifying the most viral moments and strongest short-form clip candidates from a long-form transcript. \
-For each candidate, the clip must be self-contained, starting with an extremely engaging hook within the first 3 seconds (to capture immediate attention on social feeds), \
-30-90 seconds long, and cut at clean sentence/thought boundaries. Favor highly shareable content: concrete stories, \
-strong opinions, emotional turns, surprising or counter-intuitive claims, clear payoffs, and high-energy/dramatic peaks. \
-Avoid rambling setup, context-dependent references, and pure filler. Return up to 10 candidates as JSON matching this schema: \
-{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}\n\nTranscript:\n{segments}"
+        "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection, but extract AS MANY highly viral moments as possible. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
+Clips should be 30-90 seconds long, completely self-contained, cut at clean boundaries, and deliver a massive payoff (a mind-blowing fact, hilarious joke, highly controversial opinion, or deep emotional insight). \
+Return up to 25 candidates as JSON matching exactly this schema: \
+{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}
+
+Transcript:
+{segments}"
     );
 
     let model = std::env::var("GEMINI_MODEL").unwrap_or_else(|_| "gemini-2.5-flash".to_string());
@@ -189,12 +197,16 @@ pub async fn detect_candidates_with_openai(
 ) -> Result<Vec<CandidateDraft>> {
     let segments = compact_segments(&transcript.segments);
     let prompt = format!(
-        "You are identifying the most viral moments and strongest short-form clip candidates from a long-form transcript. \
-For each candidate, the clip must be self-contained, starting with an extremely engaging hook within the first 3 seconds (to capture immediate attention on social feeds), \
-30-90 seconds long, and cut at clean sentence/thought boundaries. Favor highly shareable content: concrete stories, \
-strong opinions, emotional turns, surprising or counter-intuitive claims, clear payoffs, and high-energy/dramatic peaks. \
-Avoid rambling setup, context-dependent references, and pure filler. Return up to 10 candidates as JSON matching this schema: \
-{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}\n\nTranscript:\n{segments}"
+        "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection, but extract AS MANY highly viral moments as possible. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
+Clips should be 30-90 seconds long, completely self-contained, cut at clean boundaries, and deliver a massive payoff (a mind-blowing fact, hilarious joke, highly controversial opinion, or deep emotional insight). \
+Return up to 25 candidates as JSON matching exactly this schema: \
+{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}
+
+Transcript:
+{segments}"
     );
 
     let model = std::env::var("OPENAI_MODEL").unwrap_or_else(|_| "gpt-4o-mini".to_string());
@@ -247,12 +259,16 @@ pub async fn detect_candidates_with_openrouter(
 ) -> Result<Vec<CandidateDraft>> {
     let segments = compact_segments(&transcript.segments);
     let prompt = format!(
-        "You are identifying the most viral moments and strongest short-form clip candidates from a long-form transcript. \
-For each candidate, the clip must be self-contained, starting with an extremely engaging hook within the first 3 seconds (to capture immediate attention on social feeds), \
-30-90 seconds long, and cut at clean sentence/thought boundaries. Favor highly shareable content: concrete stories, \
-strong opinions, emotional turns, surprising or counter-intuitive claims, clear payoffs, and high-energy/dramatic peaks. \
-Avoid rambling setup, context-dependent references, and pure filler. Return up to 10 candidates as JSON matching this schema: \
-{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}\n\nTranscript:\n{segments}"
+        "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection, but extract AS MANY highly viral moments as possible. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
+Clips should be 30-90 seconds long, completely self-contained, cut at clean boundaries, and deliver a massive payoff (a mind-blowing fact, hilarious joke, highly controversial opinion, or deep emotional insight). \
+Return up to 25 candidates as JSON matching exactly this schema: \
+{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}
+
+Transcript:
+{segments}"
     );
 
     let model =
@@ -302,6 +318,71 @@ Avoid rambling setup, context-dependent references, and pure filler. Return up t
     parse_candidate_json(&text, min_duration)
 }
 
+pub async fn detect_candidates_with_groq(
+    transcript: &NormalizedTranscript,
+    api_key: &str,
+) -> Result<Vec<CandidateDraft>> {
+    let segments = compact_segments(&transcript.segments);
+    let prompt = format!(
+        "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection, but extract AS MANY highly viral moments as possible. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
+Clips should be 30-90 seconds long, completely self-contained, cut at clean boundaries, and deliver a massive payoff (a mind-blowing fact, hilarious joke, highly controversial opinion, or deep emotional insight). \
+Return up to 25 candidates as JSON matching exactly this schema: \
+{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}
+
+Transcript:
+{segments}"
+    );
+
+    let model =
+        std::env::var("GROQ_MODEL").unwrap_or_else(|_| "llama-3.3-70b-versatile".to_string());
+
+    let response = reqwest::Client::new()
+        .post("https://api.groq.com/openai/v1/chat/completions")
+        .header("Authorization", format!("Bearer {api_key}"))
+        .json(&json!({
+            "model": model,
+            "messages": [
+                {
+                    "role": "user",
+                    "content": prompt,
+                }
+            ],
+            "temperature": 0.2,
+            "response_format": {
+                "type": "json_object"
+            }
+        }))
+        .send()
+        .await
+        .context("calling Groq")?;
+
+    if !response.status().is_success() {
+        let status = response.status();
+        let body = response.text().await.unwrap_or_default();
+        return Err(anyhow!("Groq request failed ({status}): {body}"));
+    }
+
+    let res_body: ChatCompletionResponse = response
+        .json()
+        .await
+        .context("parsing Groq response")?;
+    let text = res_body
+        .choices
+        .first()
+        .map(|c| c.message.content.clone())
+        .ok_or_else(|| anyhow!("Groq response did not include choices content"))?;
+
+    let min_duration = if transcript.duration < 60.0 {
+        (transcript.duration * 0.5).max(5.0)
+    } else {
+        30.0
+    };
+    parse_candidate_json(&text, min_duration)
+}
+
 #[derive(Debug, Serialize)]
 struct ClaudeMessage<'a> {
     role: &'a str,
@@ -314,12 +395,16 @@ pub async fn detect_candidates_with_claude(
 ) -> Result<Vec<CandidateDraft>> {
     let segments = compact_segments(&transcript.segments);
     let prompt = format!(
-        "You are identifying the most viral moments and strongest short-form clip candidates from a long-form transcript. \
-For each candidate, the clip must be self-contained, starting with an extremely engaging hook within the first 3 seconds (to capture immediate attention on social feeds), \
-30-90 seconds long, and cut at clean sentence/thought boundaries. Favor highly shareable content: concrete stories, \
-strong opinions, emotional turns, surprising or counter-intuitive claims, clear payoffs, and high-energy/dramatic peaks. \
-Avoid rambling setup, context-dependent references, and pure filler. Return up to 10 candidates as JSON only: \
-{{\"candidates\":[{{\"start\":0,\"end\":0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}\n\nTranscript:\n{segments}"
+        "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection, but extract AS MANY highly viral moments as possible. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
+Clips should be 30-90 seconds long, completely self-contained, cut at clean boundaries, and deliver a massive payoff (a mind-blowing fact, hilarious joke, highly controversial opinion, or deep emotional insight). \
+Return up to 25 candidates as JSON matching exactly this schema: \
+{{\"candidates\":[{{\"start\":0.0,\"end\":0.0,\"score\":0.0,\"hook\":\"...\",\"rationale\":\"...\"}}]}}
+
+Transcript:
+{segments}"
     );
 
     let model =
@@ -381,14 +466,14 @@ pub async fn detect_candidates_with_local_llm(
 ) -> Result<Vec<CandidateDraft>> {
     let segments = compact_segments(&transcript.segments);
 
-    let system_instructions = "You are identifying the most viral moments and strongest short-form clip candidates from a long-form transcript. \
-For each candidate, the clip must be self-contained, starting with an extremely engaging hook within the first 3 seconds (to capture immediate attention on social feeds), \
-30-90 seconds long, and cut at clean sentence/thought boundaries. \
+    let system_instructions = "You are an elite, world-class social media strategist with a track record of generating viral multi-million-view Shorts, TikToks, and Reels. \
+Your sole objective is to identify the ABSOLUTE BEST, most highly-engaging, and trend-setting short-form clip candidates from the provided transcript. \
+Do NOT pick random or mediocre segments. Be ruthless in your selection. \
+Every candidate must have an insanely strong, curiosity-inducing hook in the first 3 seconds to stop the scroll. \
 CRITICAL: Each clip candidate MUST have a duration between 30 and 90 seconds (i.e. 'end' minus 'start' must be between 30.0 and 90.0). \
 Do NOT return short clips of less than 30 seconds. Combine multiple adjacent sentences to build a meaningful segment of 30-90 seconds. \
 Favor highly shareable content: concrete stories, strong opinions, emotional turns, surprising or counter-intuitive claims, clear payoffs, and high-energy/dramatic peaks. \
-Avoid rambling setup, context-dependent references, and pure filler. \
-You MUST identify and return at least 3-5 candidates (up to 10 candidates). Do not return an empty candidates list. \
+You MUST identify and return at least 3-10 candidates. Do not return an empty candidates list. \
 Ensure the 'start' and 'end' values correspond to actual timestamps in the transcript. Do not output 0.0 for start and end times.";
 
     let user_content = format!("Transcript:\n{}", segments);

src-tauri/src/models.rs

@@ -12,9 +12,11 @@ pub struct EnvironmentStatus {
     pub has_gemini_key: bool,
     pub has_openai_key: bool,
     pub has_openrouter_key: bool,
+    pub has_groq_key: bool,
     pub llm_provider: String,
     pub has_local_whisper_model: bool,
     pub has_ollama: bool,
+    pub has_ytdlp: bool,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]