One GPU. Private accounts. Characters that last.
Brew a character into images, video and film.
Apophex is a character studio that runs on one GPU. Describe a character, give it a face, approve the look, then render stills, clips and a short. Everything you make stays in your own library, and the character remembers your conversations.
- One queue, one GPU, and you can see where your job is in it
- Your characters, images and chats belong to your account alone
- Nothing expensive runs until you approve the look
Studio layout illustration. Values shown are examples.
The product
One loop, from a sentence to a short.
Most image tools stop at a picture. Apophex is built around a character that lasts: a persona you can talk to, reference images that guide the face across renders, and a pipeline that turns approved looks into turnarounds, styled poses, a trained LoRA, storyboards and video clips.
An in-app agent drives the toolchain for you. In pipeline mode it knows what is done and suggests the next step; in designer mode it interviews you and writes the persona; in character mode it simply is the character. It presents candidates. You approve.
- 01
Create
Name a character. From here on, everything you make belongs to your account alone.
- 02
Describe
Let the Design Assistant interview you, or paste references. The 6 persona sections stay hand-editable.
- 03
Imagine
Generate a look: model family, prompt, pose, LoRAs. Curate the keepers and pin the profile shot.
- 04
Approve
Approve the look once and request tweaks. This is the gate before anything expensive runs.
- 05
Build the film
Run the 8 pipeline steps one at a time, approving candidates as you go, until the clips are stitched.
Features
Everything the loop needs, in one window.
Design Assistant
Chat a character into existence. It interviews you, writes the 6 persona sections, and matches reference photos you drop in.
Imagine
Pick a model family, type a scene or let the wand write one, add a pose and LoRAs. Then clean it up: detailers, adjust, upscale, and a face model you build once and swap onto any render.
Video and film
Turn a still into motion with Wan 2.2 or MiniMax H3, or run the 8-step pipeline from character set to stitched short, approving every candidate.
In-character chat
Talk to your character, or as one of your others. It banks memories, keeps its relationships in mind, and can send a selfie mid-conversation.
Your library, everywhere
Every render lands in the right gallery. Browse it all in one place, download it as one folder, or keep a copy on your phone with the Android app.
Private accounts, one GPU
Several people share the machine. Your files, prompts and settings are yours alone; nobody else's account can open them.
Roleplay
Characters with a past, a circle and a memory.
A face is only half a character. Each one in Apophex has a backstory, a web of relationships to the others in your library that need not be mutual, and a bank of memories that grows with every conversation. Talk to them and they answer from all of it. Step into your own cast, too: choose a second character and speak as them, and the chat plays out with their history and their bond shaping every line.
Backstory
A written past the Design Assistant drafts and you keep editing, beside their look, personality, voice, outfits and world.
Relationships
Every connection points one way with a bond from 1 to 10, so Marco can be Polo's lover while Polo counts Marco a friend. 7 types, yours to rename.
Memories
The character notes what matters as you talk, timestamped and tagged by who it is about. Edit them, remove them, add your own.
Example relationships between fictional characters.
Under the hood
One GPU node, one queue.
Rendering runs in ComfyUI on one GPU node; chat, persona writing and image reading run in Ollama on the same card. Generation runs one job at a time and evicts the chat model first, so a 16GB card never fights itself. apophex.com is served by Cloudflare; the studio answers through a tunnel to the server.
ComfyUI
Z-Image, FLUX.1, SDXL and Krea 2 for stills; Wan 2.2 and MiniMax H3 for clips.
Ollama
Qwen models do the chatting, the persona writing and the image reading.
The Apophex server
One process serves the studio, the agent and the shared queue.
Your phone
The same studio in a phone browser, or the Android app with a two-way copy of your library.
Accounts and privacy
Share the GPU, not the work.
Several people can use one install. Sign-in is a session cookie, every file is resolved against whoever is asking, and nothing here sends mail: a lost password is reset by the administrator. Optional model downloads and catalog lookups contact their providers; the server operator controls storage, logs and backups.
Yours alone
- Characters, images and videos
- Projects and chat histories
- Saved prompts and favorites
- Agent model choices
- Prompt instruction edits
- Wildcard templates
Shared on the machine
- The ComfyUI install and the GPU
- Diffusion models, LoRAs and upscalers
- The pose library
- The model catalog and tag list
- The generation queue
Start brewing.
Sign in to your studio. Accounts are by invitation. If you have a code, create your account; if not, ask the administrator for one.