96 lines
4.8 KiB
Markdown
96 lines
4.8 KiB
Markdown
# Ideogram 4 — optimized local T2I
|
||
|
||
Local Ideogram 4 text-to-image workflow for the RTX 3070 server. It is
|
||
especially useful for posters, magazine covers, labels, and other images where
|
||
the prompt includes literal text.
|
||
|
||
## Baseline
|
||
|
||
- Main model: `ideogram4_int8_convrot.safetensors`
|
||
- Unconditional companion: `ideogram4_unconditional_int8_convrot.safetensors`
|
||
- Text encoder: `qwen3vl_8b_w4a8.safetensors`, type `ideogram4`
|
||
- VAE: `flux2-vae.safetensors`
|
||
- Latent/output: 768×1024 portrait
|
||
- Sampler: Euler
|
||
- Scheduler: Ideogram 4 Scheduler, `mu=0.5`, `std=1.75`
|
||
- Steps: 8
|
||
- CFG: 7.0
|
||
- CFG override: 3.0 from 70% to 100%
|
||
- Seed: 20261012 for the controlled baseline
|
||
|
||
The Fal model is retained as a comparison only; it produced more stylised,
|
||
noisier results. The hosted `IdeogramV4` API node was not used because this
|
||
server is not authenticated for Comfy API partner nodes.
|
||
|
||
## Author-inspired comparison
|
||
|
||
The model author publishes a workflow at
|
||
[`Winnougan/Ideogram_Instant_NSFW`](https://huggingface.co/Winnougan/Ideogram_Instant_NSFW).
|
||
It suggests a useful comparison branch rather than a replacement for the
|
||
local baseline:
|
||
|
||
- `Ideogram4PromptBuilderKJ` for structured prompt construction
|
||
- `ModelSamplingAuraFlow` with `shift=5.25`
|
||
- `StarSampler` with 8 steps, Euler, simple schedule, and CFG 1.0
|
||
- Fal Ideogram checkpoint with the local `flux2-vae.safetensors`
|
||
|
||
The author's graph references `Boogu\\qwen3vl_8b_int8_convrot.safetensors`, which
|
||
is not installed here, so the comparison uses the available
|
||
`qwen3vl_8b_w4a8.safetensors`. Complete settings are recorded in
|
||
[`author-inspired-settings.json`](author-inspired-settings.json).
|
||
|
||
These settings remain comparison-only until a fixed prompt and seed are tested
|
||
at 768×1024 and reviewed for typography, anatomy, texture, composition, and
|
||
runtime.
|
||
|
||
The model card also states that the fine-tune expects a structured JSON caption
|
||
with `high_level_description` and `compositional_deconstruction` fields, and
|
||
gives 10 steps with guidance scale 1.0 as its reference usage. The published
|
||
ComfyUI graph uses 8 steps, so both settings are worth measuring separately.
|
||
The card contains several unfinished placeholder sections, so its training and
|
||
fine-tuning details should not be treated as verified claims.
|
||
|
||
## Initial observations
|
||
|
||
- The 512×512 local smoke test completed successfully.
|
||
- The 768×1024 standard Ideogram 4 run took about 6 minutes at 12 steps.
|
||
- The 768×1024 Fal comparison took about 7 minutes at 12 steps.
|
||
- The 8-step Fal run completed in about 5 minutes 18 seconds, but remained
|
||
painterly and noisy.
|
||
|
||
## Quality/time test matrix
|
||
|
||
Keep the prompt and seed fixed for sampler tests. Keep all other settings fixed
|
||
when changing steps or resolution. A row is not promoted until its output is
|
||
visually reviewed for typography, anatomy, texture, and artefacts.
|
||
|
||
| ID | Resolution | Steps | Sampler | mu/std | CFG | Purpose | Status |
|
||
|---|---:|---:|---|---|---:|---|---|
|
||
| IDEO-T2I-01 | 768×1024 | 8 | Euler | 0.5/1.75 | 7 | Baseline speed/quality | Queued |
|
||
| IDEO-T2I-02 | 768×1024 | 6 | Euler | 0.5/1.75 | 7 | Faster lower bound | Queued |
|
||
| IDEO-T2I-03 | 768×1024 | 10 | Euler | 0.5/1.75 | 7 | Quality increase | Queued |
|
||
| IDEO-T2I-04 | 768×1024 | 12 | Euler | 0.5/1.75 | 7 | Existing quality reference | Complete |
|
||
| IDEO-T2I-05 | 768×1024 | 8 | Euler CFG++ | 0.5/1.75 | 7 | Sampler comparison | Queued |
|
||
| IDEO-T2I-06 | 768×1024 | 8 | Euler ancestral | 0.5/1.75 | 7 | Sampler comparison | Queued |
|
||
| IDEO-T2I-07 | 768×1024 | 8 | res_multistep | 0.5/1.75 | 7 | Sampler comparison | Queued |
|
||
| IDEO-T2I-08 | 512×768 | 8 | Euler | 0.5/1.75 | 7 | Portrait speed test | Queued |
|
||
| IDEO-T2I-09 | 1024×1024 | 8 | Euler | 0.5/1.75 | 7 | Square quality test | Queued |
|
||
| IDEO-T2I-10 | 768×1024 | 8 | Euler | 0.0/1.75 | 7 | Scheduler mu comparison | Queued |
|
||
|
||
### Author-inspired comparison matrix
|
||
|
||
| ID | Resolution | Model | Prompt path | Sampling | Purpose | Status |
|
||
|---|---:|---|---|---|---|---|
|
||
| IDEO-AUTHOR-01 | 768×1024 | Fal Ideogram | Free-form prompt | AuraFlow 5.25, StarSampler 8/Euler/simple, CFG 1 | Reproduce the author's core sampling approach | Planned |
|
||
| IDEO-AUTHOR-02 | 768×1024 | Fal Ideogram | `Ideogram4PromptBuilderKJ` | AuraFlow 5.25, StarSampler 8/Euler/simple, CFG 1 | Test structured prompt construction separately | Planned |
|
||
| IDEO-AUTHOR-03 | 768×1024 | Fal Ideogram | Structured JSON caption | Model-card reference: 10 steps, guidance 1 | Compare the documented 10-step setting | Planned |
|
||
|
||
Results belong in `results/<test-id>.png` with a matching JSON record containing
|
||
the exact prompt, seed, resolution, settings, runtime, and visual verdict.
|
||
|
||
## Reuse
|
||
|
||
Replace `REPLACE_PROMPT` in `workflow.json` and `variables.json`, then preserve
|
||
the technical baseline unless the test matrix calls for a change. For literal
|
||
text, use short exact quoted strings and explicitly forbid extra words.
|