From 4740aed9878d9865ca39adc1c9266e830019c84a Mon Sep 17 00:00:00 2001 From: Morpheus Date: Mon, 28 Sep 2026 13:00:10 +0000 Subject: [PATCH] Synchronize optimized category documentation --- optimized/MultipleRef2I/flux/README.md | 3 ++- optimized/R2V/README.md | 10 ++++++---- 2 files changed, 8 insertions(+), 5 deletions(-) diff --git a/optimized/MultipleRef2I/flux/README.md b/optimized/MultipleRef2I/flux/README.md index f2c3cf2..6770941 100644 --- a/optimized/MultipleRef2I/flux/README.md +++ b/optimized/MultipleRef2I/flux/README.md @@ -7,7 +7,8 @@ API workflow for combining two reference images with Flux.2 Klein: The workflow chains both encoded images through `ReferenceLatent` for the positive and negative conditioning branches, then renders a 1920×1080 result. Update the prompt when the references have different roles (for example, product + room, character + landscape, or object + background). -This is an API-format workflow. The tested example is stored under `fluxExperiments/multiImage/`. +This is an API-format workflow. The optimized MultipleRef2I matrix is still +untested; no result output is currently promoted for this category. ## Planned optimization matrix diff --git a/optimized/R2V/README.md b/optimized/R2V/README.md index 51ba3ca..8127530 100644 --- a/optimized/R2V/README.md +++ b/optimized/R2V/README.md @@ -1,10 +1,12 @@ # Reference-to-video workflows The optimized R2V workflows use fixed prompt, reference, seed, resolution, -duration, and CFG variables for like-for-like comparisons. Efficiency nodes -are tested first on 3-second 480p clips before longer generations. +duration, and CFG variables for like-for-like comparisons. The current +MiniMax baseline is a validated 10-second, 960×544 workflow; the earlier +3-second/480p efficiency tests remain recorded as historical comparisons. - [`minimax/`](minimax/) — MiniMax H3 R2V base and efficiency matrix -Current speed baseline: MiniMax MM-R2V-51, Turbo LoRA, 8 steps, Euler, -simple scheduler, 864×480, 3 seconds. +Current baseline: MiniMax MM-R2V-68, Turbo LoRA, SLA 0.85, Comfy Kitchen +attention, Low-VRAM Attention, and Chunk FeedForward at 8 steps, Res +Multistep, simple scheduler, 960×544, 10 seconds.