ThreadGens

ThreadGens is a local Java tool for turning Reddit-style threads or X-style posts/replies into 9:16 short-form video assets.

It can:

• read one text file and create one 1080x1920 PNG frame per line
• generate a fresh Reddit thread locally with Ollama
• generate a fresh X post and replies locally with Ollama
• generate an optional AI image for the original post through local ComfyUI / RealVisXL
• generate local voice audio with Kokoro or Piper
• render per-frame MP4 clips with FFmpeg
• stitch clips into one final MP4
• generate profile pictures and usernames for fake Reddit/X-style accounts

The default layout is built for TikTok, YouTube Shorts, and Reels-style vertical videos.

Example output

X output with OP image:

Reddit output with OP image:

Windows one-click setup

On Windows, double-click:

setup_windows.bat

The setup script will:

• check/install Java JDK 21 with winget
• check/install Python for Kokoro
• check/install Ollama with winget
• start Ollama and pull llama3.1:8b
• download Piper TTS for Windows
• download the en_US-lessac-medium Piper voice
• create an isolated .venv-kokoro environment
• install Kokoro TTS requirements
• build ThreadGens into out/
• check profile-generator support

After setup finishes, double-click:

run_ai_windows.bat

That runner asks for platform/thread style, TTS engine, prompt text, count, optional OP image generation, and optional final video output.

For Reddit:

post title
post body

For X:

optional hidden reply style
visible original X post text

Normal and batch Windows runners keep Ollama loaded by default so repeated generations are faster. They only unload Ollama if you choose that option.

Default outputs:

output/script/generated_comments.txt
output/*.png
output/images/*.png
output/cache/images/*.txt
output/audio/*.wav
output/video/*.mp4
output/video/final.mp4

Build manually

javac -d out src/redditTxtToImg/*.java

Or with Gradle:

gradle build

Recommended CLI entrypoint

Use CheckedRunner for scripts and normal command-line runs:

java -cp out redditTxtToImg.CheckedRunner data/comments.txt output

CheckedRunner delegates to the selected renderer, then verifies that the expected fresh PNG/WAV/MP4 files were actually created. This prevents failed runs from looking successful.

Use OpImageVideoSafeRunner when OP image generation and video creation are both enabled. It delays MP4 creation until after the OP image has been overlaid onto the original post PNG.

java -cp out redditTxtToImg.OpImageVideoSafeRunner --platform x --auto --topic "I just saw something weird and I need someone else to explain it." --count 10 --image-mode comfyui --tts kokoro --tts-command .venv-kokoro/Scripts/python.exe --voice af_heart --video --concat-video

Select a platform with --platform:

java -cp out redditTxtToImg.CheckedRunner data/comments.txt output --platform reddit
java -cp out redditTxtToImg.CheckedRunner data/comments.txt output --platform x

Supported values:

• reddit
• x
• twitter as an alias for x

The older renderer entrypoints still exist:

java -cp out redditTxtToImg.RedditScreenshotGenerator data/comments.txt output
java -cp out redditTxtToImg.XThreadGenerator data/comments.txt output

Use them only when you want raw renderer behavior without checked artifact validation.

GUI

java -cp out redditTxtToImg.GuiApp

The GUI lets you choose an input text file, choose an output folder, and generate images. It uses the checked runner so missing output files show as failures instead of false success.

CLI options

Basic Reddit image generation:

java -cp out redditTxtToImg.CheckedRunner data/comments.txt output --platform reddit --count 3 --prefix sample --style light --shuffle --center

Basic X image generation:

java -cp out redditTxtToImg.CheckedRunner data/comments.txt output --platform x --count 3 --prefix xsample --top

Image options:

• --platform reddit|x chooses the screenshot/thread style. Default: reddit.
• --count N limits how many frames are generated.
• --prefix NAME changes output file names. Files are named like 0NAME.png, 1NAME.png, etc.
• --style dark|light chooses a Reddit template from templates/.
• --shuffle shuffles input lines before rendering.
• --center centers Reddit comments in the vertical card.
• --top keeps text closer to the top of the card.
• --no-watermark hides the small watermark.
• --gui opens the GUI.

Local AI text generation

Make sure Ollama is running locally and the model is already pulled:

ollama pull llama3.1:8b

Reddit auto-generation:

java -cp out redditTxtToImg.CheckedRunner --platform reddit --auto --post-title "Finish this story in the comments" --topic "I found a locked box behind my dryer." --count 10 --llm-model llama3.1:8b

X auto-generation:

java -cp out redditTxtToImg.CheckedRunner --platform x --auto --post-title "Wrong answers only" --topic "Why is there a shopping cart in my living room?" --count 10 --llm-model llama3.1:8b

For X, --post-title is not displayed. It is only a hidden reply style/instruction. The visible X post is --topic.

X example with no hidden style:

java -cp out redditTxtToImg.CheckedRunner --platform x --auto --topic "I just saw something weird and I need someone else to explain it." --count 5 --llm-model llama3.1:8b

Local AI options:

• --auto asks the local LLM to generate a fresh script before rendering.
• --post-title TEXT sets the visible Reddit title, or hidden X reply style.
• --topic TEXT sets the Reddit post body/story, or visible X post text.
• --llm-model MODEL sets the Ollama model name. Default: llama3.1:8b.
• --llm-url URL sets the Ollama generate endpoint. Default: http://localhost:11434/api/generate.
• --script-out FILE changes where generated text is saved.
• --keep-ollama-loaded keeps Ollama loaded after script generation. The Windows runners and defaults.txt already use keep-loaded behavior by default.

OP image generation

OP image generation is optional and off by default for smoke tests and text-only runs.

Two modes are supported:

• --image-mode comfyui generates an OP image locally through ComfyUI.
• --image-mode local --op-image path/to/image.png uses an existing local image for testing the overlay.

ComfyUI mode uses Ollama to expand the OP title/body into an SDXL prompt, then sends that prompt to local ComfyUI. The default checkpoint is:

RealVisXL_V5.0_fp32.safetensors

ComfyUI must already be running at:

http://127.0.0.1:8188

Reddit OP image example:

java -cp out redditTxtToImg.OpImageVideoSafeRunner --platform reddit --auto \
  --post-title "Finish this story in the comments" \
  --topic "weird everyday stories" \
  --count 10 \
  --image-mode comfyui \
  --tts kokoro \
  --tts-command .venv-kokoro/Scripts/python.exe \
  --voice af_heart \
  --video \
  --concat-video

X OP image example:

java -cp out redditTxtToImg.OpImageVideoSafeRunner --platform x --auto \
  --topic "I just saw something weird and I need someone else to explain it." \
  --count 10 \
  --image-mode comfyui \
  --tts kokoro \
  --tts-command .venv-kokoro/Scripts/python.exe \
  --voice af_heart \
  --video \
  --concat-video

OP image options:

• --image-mode none|local|comfyui controls OP image behavior.
• --op-image FILE supplies a local test image when using --image-mode local.
• --image-dir DIR changes where generated OP images are saved.
• --image-cache-dir DIR changes where generated prompt cache files are saved.
• --comfy-url URL changes the ComfyUI server URL.
• --image-checkpoint NAME changes the ComfyUI checkpoint name.
• --image-width N and --image-height N change generated image dimensions.
• --image-steps N, --image-cfg N, --image-sampler NAME, and --image-scheduler NAME tune ComfyUI generation.
• --image-negative TEXT changes the negative prompt.
• --image-timeout SECONDS changes the ComfyUI polling timeout.

Local TTS

ThreadGens supports:

• --tts kokoro for Kokoro voices such as af_heart, af_bella, am_adam, etc.
• --tts piper for Piper .onnx voice models.
• --tts none to skip audio.

Kokoro Reddit example:

java -cp out redditTxtToImg.CheckedRunner --platform reddit --auto \
  --post-title "Finish this story in the comments" \
  --topic "I found a second phone hidden in my car." \
  --count 10 \
  --tts kokoro \
  --tts-command .venv-kokoro/Scripts/python.exe \
  --voice af_heart

Kokoro X example:

java -cp out redditTxtToImg.CheckedRunner --platform x --auto \
  --post-title "Wrong answers only" \
  --topic "Why is there a shopping cart in my living room?" \
  --count 10 \
  --tts kokoro \
  --tts-command .venv-kokoro/Scripts/python.exe \
  --voice af_heart

Piper example:

java -cp out redditTxtToImg.CheckedRunner --platform reddit --auto \
  --post-title "Finish this story in the comments" \
  --topic "creepy small town stories" \
  --count 10 \
  --tts piper \
  --voice voices/en_US-lessac-medium.onnx

TTS options:

• --tts none|piper|kokoro enables or disables voice generation.
• --voice NAME_OR_PATH sets the Kokoro voice name or Piper voice path/name.
• --voice-dir DIR sets the Piper voice directory.
• --list-voices lists local Piper voices and common Kokoro voices.
• --tts-command CMD changes the executable used by the selected TTS engine.
• --audio-dir DIR changes where WAV files are saved.
• --tts-timeout SECONDS changes the per-line TTS timeout.

Video generation

Video generation requires FFmpeg and FFprobe on PATH, or a direct path passed with --video-command.

java -cp out redditTxtToImg.CheckedRunner --platform x --auto \
  --post-title "Wrong answers only" \
  --topic "Why is there a shopping cart in my living room?" \
  --count 10 \
  --tts kokoro \
  --tts-command .venv-kokoro/Scripts/python.exe \
  --voice af_heart \
  --video \
  --concat-video

When --image-mode comfyui and video are both enabled, prefer OpImageVideoSafeRunner so the final MP4 includes the OP image overlay.

Video options:

• --video renders one MP4 clip per PNG/WAV pair.
• --concat-video also stitches those clips into one final MP4.
• --video-dir DIR changes where clips and final videos are saved.
• --video-command CMD changes the FFmpeg executable path.
• --fps 30 sets video FPS.
• --video-timeout SECONDS changes the FFmpeg timeout.
• --final-video NAME.mp4 changes the stitched final video name.

Batch video creation

Create or edit:

data/batch_videos.txt

Format uses two non-empty lines per video.

For Reddit:

line 1 = post title
line 2 = post body
line 3 = next post title
line 4 = next post body

For X:

line 1 = hidden reply style
line 2 = visible X post text
line 3 = next hidden reply style
line 4 = next visible X post text

Then run:

batch_create_videos_windows.bat

The batch runner asks whether to generate OP images for each video. It keeps Ollama loaded between videos by default.

Batch outputs go under:

output/batch_videos/<platform>_<timestamp>/

Final MP4 copies are collected in:

output/batch_videos/<platform>_<timestamp>/final_videos/

Profile pictures and usernames

Run:

generate_profiles_windows.bat

Choose either:

1. AI face/selfie profile pictures through local ComfyUI.
2. Fast procedural fallback avatars with no AI model.

Direct procedural mode:

python tools/generate_profiles.py --count 100 --size 256

Direct ComfyUI mode:

python tools/generate_comfy_profiles.py --count 25 --size 512 --comfy-url http://127.0.0.1:8188

Generated profile pictures go to:

assets/pfp/

Generated usernames go to:

data/author_names.txt

Files

• src/redditTxtToImg/ contains the Java source.
• src/redditTxtToImg/CheckedRunner.java is the safe CLI/script entrypoint.
• src/redditTxtToImg/OpImageVideoSafeRunner.java is the safe entrypoint for OP-image plus video runs.
• src/redditTxtToImg/RedditScreenshotGenerator.java renders Reddit-style frames.
• src/redditTxtToImg/XThreadGenerator.java renders X-style frames.
• src/redditTxtToImg/LocalLlmTextGenerator.java handles Reddit Ollama generation.
• src/redditTxtToImg/XLocalLlmTextGenerator.java handles X Ollama generation.
• src/redditTxtToImg/OpImagePipeline.java coordinates OP image prompting, generation, and overlay.
• src/redditTxtToImg/ComfyUiImageGenerator.java calls the local ComfyUI API.
• src/redditTxtToImg/OpImagePromptGenerator.java asks Ollama to write the image prompt.
• data/author_names.txt contains sample names.
• data/comments.txt contains sample input lines.
• defaults.txt contains default render and local pipeline settings.
• templates/ contains simple Reddit color templates.
• assets/ contains optional images.
• docs/x-op-image-example.png shows the X README output example.
• docs/reddit-op-image-example.png shows the Reddit README output example.
• tools/kokoro_tts.py is the Kokoro helper used by Java.
• tools/generate_profiles.py creates procedural avatars and usernames.
• tools/generate_comfy_profiles.py creates ComfyUI profile images and usernames.
• tools/batch_create_videos.ps1 powers batch video creation.
• tools/run_ai_windows.ps1 powers the normal Windows AI runner.
• setup_windows.bat launches Windows setup.
• run_ai_windows.bat launches the normal local AI pipeline.

Runnable jar

gradle jar
java -jar build/libs/ThreadGens-0.4.2-op-image-generation.jar data/comments.txt output

Runnable jar with local AI:

java -jar build/libs/ThreadGens-0.4.2-op-image-generation.jar --platform x --auto --post-title "Wrong answers only" --topic "Why is there a shopping cart in my living room?" --count 10 --tts kokoro --voice af_heart