diff --git a/src/config.rs b/src/config.rs
index 25c6300..65f5138 100644
--- a/src/config.rs
+++ b/src/config.rs
@@ -145,8 +145,8 @@ impl Config {
     /// Load configuration from a TOML file.
     pub fn from_file(path: impl AsRef<std::path::Path>) -> crate::error::Result<Self> {
         let content = std::fs::read_to_string(path)?;
-        let config: Config = toml::from_str(&content)
-            .map_err(|e| crate::error::Error::Config(e.to_string()))?;
+        let config: Config =
+            toml::from_str(&content).map_err(|e| crate::error::Error::Config(e.to_string()))?;
         Ok(config)
     }
 
diff --git a/src/job/mod.rs b/src/job/mod.rs
index 562106c..bdeca50 100644
--- a/src/job/mod.rs
+++ b/src/job/mod.rs
@@ -1,8 +1,467 @@
 //! Background job processing module.
+//!
+//! Handles the complete pipeline:
+//! 1. Fetching assets from Immich for a specific person
+//! 2. Downloading and processing images in parallel
+//! 3. Cropping faces using bounding boxes
+//! 4. Compiling processed images into a timelapse video
 
-// TODO: Implement job processor
-// This will handle:
-// - Fetching assets from Immich
-// - Parallel face processing
-// - Progress reporting
-// - Cancellation handling
+use crate::config::Config;
+use crate::error::{Error, Result};
+use crate::face_processing::{AssetResult, BoundingBox, ProcessedFace};
+use crate::immich_api::{Asset, FaceData, ImmichClient};
+use crate::video::compile_timelapse;
+use crate::web::{AppState, JobStatus, Progress};
+
+use image::imageops::FilterType;
+use image::{DynamicImage, GenericImageView, ImageFormat};
+use std::io::Cursor;
+use std::path::{Path, PathBuf};
+use std::sync::atomic::{AtomicU32, Ordering};
+use std::sync::Arc;
+use tokio::sync::{oneshot, Semaphore};
+
+/// Parameters for starting a processing job.
+#[derive(Debug, Clone)]
+pub struct JobParams {
+    pub person_id: String,
+    pub date_from: Option<String>,
+    pub date_to: Option<String>,
+}
+
+/// Run the complete processing pipeline.
+///
+/// This is the main entry point for background job processing.
+/// It handles progress reporting and cancellation.
+pub async fn run_job(state: AppState, params: JobParams, cancel_rx: oneshot::Receiver<()>) {
+    let result = run_job_inner(state.clone(), params, cancel_rx).await;
+
+    match result {
+        Ok(output_path) => {
+            state
+                .update_progress(Progress {
+                    status: JobStatus::Completed,
+                    completed: 0,
+                    total: 0,
+                    message: Some(format!("Video saved to: {}", output_path.display())),
+                })
+                .await;
+        }
+        Err(Error::Cancelled) => {
+            state
+                .update_progress(Progress {
+                    status: JobStatus::Cancelled,
+                    completed: 0,
+                    total: 0,
+                    message: Some("Job cancelled by user".to_string()),
+                })
+                .await;
+        }
+        Err(e) => {
+            tracing::error!("Job failed: {}", e);
+            state
+                .update_progress(Progress {
+                    status: JobStatus::Error(e.to_string()),
+                    completed: 0,
+                    total: 0,
+                    message: Some(e.to_string()),
+                })
+                .await;
+        }
+    }
+}
+
+/// Inner implementation that returns Result for easier error handling.
+async fn run_job_inner(
+    state: AppState,
+    params: JobParams,
+    mut cancel_rx: oneshot::Receiver<()>,
+) -> Result<PathBuf> {
+    let config = state.config.read().await.clone();
+
+    // Create output directories
+    let images_dir = config.output_dir.join("images");
+    tokio::fs::create_dir_all(&images_dir).await?;
+
+    // Create Immich client
+    let client = ImmichClient::new(&config.api)?;
+
+    // Update progress: fetching assets
+    state
+        .update_progress(Progress {
+            status: JobStatus::Running,
+            completed: 0,
+            total: 0,
+            message: Some("Fetching assets from Immich...".to_string()),
+        })
+        .await;
+
+    // Check for cancellation
+    if cancel_rx.try_recv().is_ok() {
+        return Err(Error::Cancelled);
+    }
+
+    // Fetch all assets for this person
+    let assets = client
+        .get_assets_with_person(
+            &params.person_id,
+            params.date_from.as_deref(),
+            params.date_to.as_deref(),
+        )
+        .await?;
+
+    if assets.is_empty() {
+        return Err(Error::ImageProcessing(
+            "No assets found for this person".to_string(),
+        ));
+    }
+
+    tracing::info!("Found {} assets to process", assets.len());
+
+    // Filter assets that have face data for the target person
+    let assets_with_faces: Vec<(Asset, FaceData)> = assets
+        .into_iter()
+        .filter_map(|asset| {
+            let face = find_face_for_person(&asset, &params.person_id)?;
+            Some((asset, face))
+        })
+        .collect();
+
+    if assets_with_faces.is_empty() {
+        return Err(Error::ImageProcessing(
+            "No assets with face data found".to_string(),
+        ));
+    }
+
+    let total = assets_with_faces.len() as u32;
+    tracing::info!("{} assets have face data", total);
+
+    // Update progress with total count
+    state
+        .update_progress(Progress {
+            status: JobStatus::Running,
+            completed: 0,
+            total,
+            message: Some(format!("Processing {} images...", total)),
+        })
+        .await;
+
+    // Process images in parallel with concurrency limit
+    let completed = Arc::new(AtomicU32::new(0));
+    let semaphore = Arc::new(Semaphore::new(config.processing.max_workers));
+    let client = Arc::new(client);
+    let config = Arc::new(config);
+    let images_dir = Arc::new(images_dir);
+
+    let mut handles = Vec::with_capacity(assets_with_faces.len());
+
+    for (asset, face_data) in assets_with_faces {
+        let permit = semaphore.clone().acquire_owned().await.unwrap();
+        let client = client.clone();
+        let config = config.clone();
+        let images_dir = images_dir.clone();
+        let completed = completed.clone();
+        let state = state.clone();
+
+        let handle = tokio::spawn(async move {
+            let result = process_single_asset(&client, &config, &asset, &face_data, &images_dir).await;
+
+            // Update progress
+            let done = completed.fetch_add(1, Ordering::SeqCst) + 1;
+            state
+                .update_progress(Progress {
+                    status: JobStatus::Running,
+                    completed: done,
+                    total,
+                    message: Some(format!("Processing images... ({}/{})", done, total)),
+                })
+                .await;
+
+            drop(permit);
+            result
+        });
+
+        handles.push(handle);
+    }
+
+    // Check for cancellation periodically while waiting
+    let mut results = Vec::with_capacity(handles.len());
+    for handle in handles {
+        tokio::select! {
+            _ = &mut cancel_rx => {
+                return Err(Error::Cancelled);
+            }
+            result = handle => {
+                results.push(result.unwrap_or(AssetResult::Error {
+                    asset_id: "unknown".to_string(),
+                    error: "Task panicked".to_string(),
+                }));
+            }
+        }
+    }
+
+    // Count results
+    let successful = results
+        .iter()
+        .filter(|r| matches!(r, AssetResult::Success(_)))
+        .count();
+    let skipped = results
+        .iter()
+        .filter(|r| matches!(r, AssetResult::Skipped { .. }))
+        .count();
+    let errors = results
+        .iter()
+        .filter(|r| matches!(r, AssetResult::Error { .. }))
+        .count();
+
+    tracing::info!(
+        "Processing complete: {} successful, {} skipped, {} errors",
+        successful,
+        skipped,
+        errors
+    );
+
+    if successful == 0 {
+        return Err(Error::ImageProcessing(
+            "No images were successfully processed".to_string(),
+        ));
+    }
+
+    // Check for cancellation before video compilation
+    if cancel_rx.try_recv().is_ok() {
+        return Err(Error::Cancelled);
+    }
+
+    // Compile video if enabled
+    if config.video.enabled {
+        state
+            .update_progress(Progress {
+                status: JobStatus::CompilingVideo,
+                completed: 0,
+                total: successful as u32,
+                message: Some("Compiling video...".to_string()),
+            })
+            .await;
+
+        let output_video = config.output_dir.join("timelapse.mp4");
+        let state_clone = state.clone();
+
+        compile_timelapse(&images_dir, &output_video, &config.video, move |frame, total| {
+            // Note: This callback is sync, so we can't easily update state here.
+            // The FFmpeg wrapper already handles this internally.
+            tracing::debug!("Video progress: {}/{}", frame, total);
+            let _ = (&state_clone, frame, total); // Suppress unused warning
+        })
+        .await?;
+
+        Ok(output_video)
+    } else {
+        Ok(images_dir.to_path_buf())
+    }
+}
+
+/// Find the face data for a specific person in an asset.
+fn find_face_for_person(asset: &Asset, person_id: &str) -> Option<FaceData> {
+    let people = asset.people.as_ref()?;
+
+    for person in people {
+        if person.id == person_id {
+            if let Some(faces) = &person.faces {
+                // Return the first face (usually there's only one per person per image)
+                return faces.first().cloned();
+            }
+        }
+    }
+
+    None
+}
+
+/// Process a single asset: download, crop face, save.
+async fn process_single_asset(
+    client: &ImmichClient,
+    config: &Config,
+    asset: &Asset,
+    face_data: &FaceData,
+    output_dir: &Path,
+) -> AssetResult {
+    let asset_id = &asset.id;
+
+    // Check face resolution
+    let bbox = BoundingBox {
+        x1: face_data.bounding_box_x1,
+        y1: face_data.bounding_box_y1,
+        x2: face_data.bounding_box_x2,
+        y2: face_data.bounding_box_y2,
+    };
+
+    let face_width = bbox.width() * face_data.image_width as f32;
+    let face_height = bbox.height() * face_data.image_height as f32;
+    let face_size = face_width.min(face_height) as u32;
+
+    if face_size < config.processing.face_resolution_threshold {
+        return AssetResult::Skipped {
+            asset_id: asset_id.clone(),
+            reason: format!(
+                "Face too small: {}px (threshold: {}px)",
+                face_size, config.processing.face_resolution_threshold
+            ),
+        };
+    }
+
+    // Download image
+    let image_bytes = match client.download_asset(asset_id).await {
+        Ok(bytes) => bytes,
+        Err(e) => {
+            return AssetResult::Error {
+                asset_id: asset_id.clone(),
+                error: format!("Download failed: {}", e),
+            };
+        }
+    };
+
+    // Decode image
+    let img = match image::load_from_memory(&image_bytes) {
+        Ok(img) => img,
+        Err(e) => {
+            return AssetResult::Error {
+                asset_id: asset_id.clone(),
+                error: format!("Failed to decode image: {}", e),
+            };
+        }
+    };
+
+    // Crop and resize face
+    let cropped = match crop_face(&img, face_data, config.processing.resize_size) {
+        Ok(cropped) => cropped,
+        Err(e) => {
+            return AssetResult::Error {
+                asset_id: asset_id.clone(),
+                error: format!("Failed to crop face: {}", e),
+            };
+        }
+    };
+
+    // Generate timestamp-based filename for sorting
+    let timestamp = asset
+        .file_created_at
+        .as_ref()
+        .or(asset.local_date_time.as_ref())
+        .cloned()
+        .unwrap_or_else(|| asset_id.clone());
+
+    // Sanitize timestamp for filename (replace colons and other invalid chars)
+    let safe_timestamp: String = timestamp
+        .chars()
+        .map(|c| if c.is_alphanumeric() || c == '-' || c == '_' { c } else { '_' })
+        .collect();
+
+    let output_path = output_dir.join(format!("{}_{}.jpg", safe_timestamp, asset_id));
+
+    // Encode and save
+    let mut buffer = Cursor::new(Vec::new());
+    if let Err(e) = cropped.write_to(&mut buffer, ImageFormat::Jpeg) {
+        return AssetResult::Error {
+            asset_id: asset_id.clone(),
+            error: format!("Failed to encode image: {}", e),
+        };
+    }
+
+    if let Err(e) = tokio::fs::write(&output_path, buffer.into_inner()).await {
+        return AssetResult::Error {
+            asset_id: asset_id.clone(),
+            error: format!("Failed to save image: {}", e),
+        };
+    }
+
+    AssetResult::Success(ProcessedFace {
+        image_data: Vec::new(), // We saved to file, so don't keep in memory
+        asset_id: asset_id.clone(),
+        timestamp,
+    })
+}
+
+/// Crop and resize the face from an image using bounding box.
+///
+/// This is a simplified version that just uses the bounding box.
+/// A full implementation would use facial landmarks for alignment.
+fn crop_face(img: &DynamicImage, face_data: &FaceData, output_size: u32) -> Result<DynamicImage> {
+    let (img_width, img_height) = img.dimensions();
+
+    // Convert normalized bounding box to pixel coordinates
+    let x1 = (face_data.bounding_box_x1 * img_width as f32) as u32;
+    let y1 = (face_data.bounding_box_y1 * img_height as f32) as u32;
+    let x2 = (face_data.bounding_box_x2 * img_width as f32) as u32;
+    let y2 = (face_data.bounding_box_y2 * img_height as f32) as u32;
+
+    let face_width = x2.saturating_sub(x1);
+    let face_height = y2.saturating_sub(y1);
+
+    if face_width == 0 || face_height == 0 {
+        return Err(Error::ImageProcessing("Invalid face bounding box".to_string()));
+    }
+
+    // Expand the crop area to include some context around the face
+    // and make it square for consistent output
+    let face_size = face_width.max(face_height);
+    let padding = face_size / 2; // 50% padding on each side
+    let crop_size = face_size + padding * 2;
+
+    // Calculate center of face
+    let center_x = (x1 + x2) / 2;
+    let center_y = (y1 + y2) / 2;
+
+    // Calculate crop bounds, clamped to image dimensions
+    let crop_x1 = center_x.saturating_sub(crop_size / 2).min(img_width.saturating_sub(crop_size));
+    let crop_y1 = center_y.saturating_sub(crop_size / 2).min(img_height.saturating_sub(crop_size));
+
+    // Ensure we don't exceed image bounds
+    let actual_crop_size = crop_size.min(img_width - crop_x1).min(img_height - crop_y1);
+
+    if actual_crop_size < 10 {
+        return Err(Error::ImageProcessing("Crop area too small".to_string()));
+    }
+
+    // Crop the face region
+    let cropped = img.crop_imm(crop_x1, crop_y1, actual_crop_size, actual_crop_size);
+
+    // Resize to output size
+    let resized = cropped.resize_exact(output_size, output_size, FilterType::Lanczos3);
+
+    Ok(resized)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_find_face_for_person() {
+        use crate::immich_api::PersonWithFaces;
+
+        let asset = Asset {
+            id: "asset1".to_string(),
+            device_asset_id: None,
+            original_file_name: None,
+            file_created_at: Some("2024-01-15T10:30:00Z".to_string()),
+            local_date_time: None,
+            people: Some(vec![PersonWithFaces {
+                id: "person1".to_string(),
+                name: Some("Test Person".to_string()),
+                faces: Some(vec![FaceData {
+                    bounding_box_x1: 0.2,
+                    bounding_box_y1: 0.1,
+                    bounding_box_x2: 0.5,
+                    bounding_box_y2: 0.6,
+                    image_width: 1920,
+                    image_height: 1080,
+                }]),
+            }]),
+        };
+
+        let face = find_face_for_person(&asset, "person1");
+        assert!(face.is_some());
+
+        let face = find_face_for_person(&asset, "person2");
+        assert!(face.is_none());
+    }
+}
diff --git a/src/web/handlers.rs b/src/web/handlers.rs
index d9bfc93..3f2a99b 100644
--- a/src/web/handlers.rs
+++ b/src/web/handlers.rs
@@ -1,6 +1,7 @@
 //! HTTP route handlers.
 
 use crate::immich_api::ImmichClient;
+use crate::job::{run_job, JobParams};
 use crate::web::state::{AppState, JobStatus, Progress};
 use axum::{
     extract::State,
@@ -169,11 +170,9 @@ async fn start_processing(
         .await;
 
     // Create cancellation channel
-    let (cancel_tx, _cancel_rx) = tokio::sync::oneshot::channel();
+    let (cancel_tx, cancel_rx) = tokio::sync::oneshot::channel();
     state.set_cancel_sender(cancel_tx).await;
 
-    // TODO: Spawn actual processing task
-    // For now, just return success
     tracing::info!(
         "Starting processing for person {} (date range: {:?} - {:?})",
         request.person_id,
@@ -181,6 +180,18 @@ async fn start_processing(
         request.date_to
     );
 
+    // Spawn the processing job in the background
+    let job_params = JobParams {
+        person_id: request.person_id,
+        date_from: request.date_from,
+        date_to: request.date_to,
+    };
+
+    let job_state = state.clone();
+    tokio::spawn(async move {
+        run_job(job_state, job_params, cancel_rx).await;
+    });
+
     Ok(Json(StartResponse {
         success: true,
         message: "Processing started".to_string(),