3D + AI controllable workflows
Process videos and node-level walkthroughs for serious builders.
Enterprise workflows · PoC Reproducible
Reproducible
AI production
for teams
ComfyUI pipelines, local bf16/fp16 operation, API export, and batch production. I build systems that automate creative work — not single images on demand.
Core Differentiator
Atelier Operations — 4 Agents
I don't only make works — I build systems that let AI run production, distribution, and opportunity scouting for me.
Creator literacy · Behind the scenes
How we work in public
AI tech news, translated into craft
Overseas signals → English explainers, posts, and reusable prompts.
Fantasy characters · cosmic civilization
Worldboard thinking plus character-consistent generation demos.
GeekSpell · AI literacy series
Codex / Grok / prompt constraints & reproducibility — YouTube · GeekSpell Studio.
YouTubeXSubstackGumroadProduct Hunt
Works Mothership
Perception · Series · Archive
High-VRAM Workflow Pack
Production workflows — 48–96 GB class
Adapted from the author's ComfyUI library. Larger tiles, tiling off, PurgeVRAM bypassed. Local execution only — verify model paths.
48–96 GB · 6 files
Min VRAM: 24 GB+ · image-edit
Boogu image edit (2 refs)
Default Eliora workflow — TextEncodeBooguEdit, 2 LoadImage slots. User-tested JSON.
Source: Boogu · Eliora verified 2026-07-09
Download JSON Min VRAM: 48 GB+ · product-imaging
SeedVR2 billion-pixel upscale
Tiled upscale tuned for 48–96GB VRAM — larger tiles, PurgeVRAM bypassed.
Source: 像素幻想Lab (adapted by Miragea · Eliora)
Download JSON Min VRAM: 32 GB+ · ad-workflows
Flux2 Klein — full image edit suite
Multi-route Klein edit flow — 1536→2048+ longest edge, tiling off where possible.
Source: TrueV2 collection (adapted by Miragea · Eliora)
Download JSON Min VRAM: 80 GB+ · video-pipeline
Wan 2.2 — 14B image-to-video
14B I2V — higher frame size preset for 80GB+ class GPUs. Swap fp8→fp16 weights if you have headroom.
Source: Comfy-Org Wan 2.2 repack (adapted by Miragea · Eliora)
Download JSON Min VRAM: 48 GB+ · product-imaging
Dual-sampler wash + ultra upscale
Fast wash + upscale path for large VRAM — fewer VRAM purge passes.
Source: 像素幻想Lab (adapted by Miragea · Eliora)
Download JSON Min VRAM: 48 GB+ · character-consistency
Flux2 Klein T2I — portrait LoRA
Portrait-optimized Klein T2I — latent size bumped for 48GB+.
Source: flux2 klein portrait (adapted by Miragea · Eliora)
Download JSONEliora AI Studio
Seven Production Lines
Reproducibility, control, batch production, API — not single images.
◈
Product Imaging
◇
Ad Workflows
◎
Video Pipeline
◉
Character Consistency
⬡
ComfyUI API
▣
Enterprise
◆
Workflow Market
Continuing Series
Who Will Govern the Future?
EP.01 · 3:12
The Question
Who decides when machines decide?
EP.02 · 2:58
The Archive
Memory, power, and what we keep.
EP.03 · 3:05
The Pipeline
From prompt to policy — who governs?
EP.04
The Agent
When AI advances the next job alone.
XYouTubePatreonGumroadSubstackProduct Hunt
Enterprise & Projects
Serious work inquiries
PoC, ComfyUI automation, workflow builds, consulting. Use the fixed inquiry form — replies go to your email.
Open inquiry form