Quick Answer
Boogu/Boogu-Image-0.1-Edit (Boogu) is the best image editing model to run locally on a 16 GB GPU as of September 2026, picked as the newest release of the current open-weight generation (published 2026-06-16, 10.3B parameters, Apache-2.0) because no public benchmark yet scores these candidates against each other [5]. In order, the ranking is Boogu/Boogu-Image-0.1-Edit (Boogu), FireRedTeam/FireRed-Image-Edit-1.1 (FireRedTeam), and Qwen/Qwen-Image-Edit-2511 (Qwen) [5][3].
Key Takeaways
- Boogu/Boogu-Image-0.1-Edit (Boogu) takes #1 as the newest release in the field, first published on 2026-06-16 [5]; no public benchmark yet scores any ranked candidate, so the ordering is newest release first, then downloads.
- Boogu/Boogu-Image-0.1-Edit is the easiest fit for 16 GB: 4-bit weights come to roughly 5.2 GB, estimated (params x 0.5 bytes) from 10.3B parameters [5], versus a 38.5 GB BF16 checkpoint [5], with the system documented in the Boogu-Image-0.1 paper (arXiv 2607.13125) [6].
- FireRedTeam/FireRed-Image-Edit-1.1 (FireRedTeam) ranks second on release date — 2026-03-02, superseding FireRed-Image-Edit-1.0 [3] — but 20.4B parameters [3] put 4-bit weights near 10.2 GB, estimated (params x 0.5 bytes), before activations and context on a 16 GB card.
- Qwen/Qwen-Image-Edit-2511 (Qwen) sits third only because its 2025-12-17 release is the oldest [1]; 323,535 downloads in the last 30 days make it the most downloaded candidate over 8,352 and 1,119 [1][3][5], a tiebreak only, and the Qwen-Image Technical Report (arXiv 2508.02324) covers the family [2].
- All three candidates ship under Apache-2.0, permitting commercial local deployment across the board [1][3][5].
- No public leaderboard scores Boogu/Boogu-Image-0.1-Edit [5], FireRedTeam/FireRed-Image-Edit-1.1 [3], or Qwen/Qwen-Image-Edit-2511 [1] against each other, so treat any model-card performance figures as self-reported until a shared benchmark appears.
How do the best open-weight image editing models for a 16 GB GPU compare on specs and memory?
| Model | Org | Params | Quant/VRAM | Released (date) | License | Key benchmark (date) |
|---|---|---|---|---|---|---|
| Boogu/Boogu-Image-0.1-Edit | Boogu | 10.3B [5] | BF16 38.5 GB [5]; 4-bit ≈ 5.2 GB estimated (10.3B [5] × 0.5 bytes) | 2026-06-16 [5] | Apache-2.0 [5] | No public benchmark yet |
| FireRedTeam/FireRed-Image-Edit-1.1 | FireRedTeam | 20.4B [3] | BF16 57.7 GB [3]; 4-bit ≈ 10.2 GB estimated (20.4B [3] × 0.5 bytes) | 2026-03-02 [3] | Apache-2.0 [3] | No public benchmark yet |
| Qwen/Qwen-Image-Edit-2511 | Qwen | 20.4B [1] | BF16 57.7 GB [1]; 4-bit ≈ 10.2 GB estimated (20.4B [1] × 0.5 bytes) | 2025-12-17 [1] | Apache-2.0 [1] | No public benchmark yet |
What is the best image editing model for a 16 GB GPU in 2026?
1. Boogu/Boogu-Image-0.1-Edit
Boogu/Boogu-Image-0.1-Edit from Boogu takes the top slot because it is the newest release among the ranked editors, first published on Hugging Face on 2026-06-16 [5], ahead of FireRedTeam/FireRed-Image-Edit-1.1 (2026-03-02) [3] and Qwen/Qwen-Image-Edit-2511 (2025-12-17) [1]. Since none of the ranked models has a public benchmark score, the ordering runs newest release first, then downloads; Boogu-Image-0.1-Edit itself has no public benchmark yet.
Key specifications are 10.3B parameters, BF16 weights at 38.5 GB, and an Apache-2.0 license [5]. That 38.5 GB BF16 checkpoint [5] will not fit in 16 GB of VRAM, so plan on quantization: a 4-bit copy runs about 5.2 GB, estimated (params x 0.5 bytes) from the 10.3B parameter count [5], leaving headroom on a 16 GB GPU for context and runtime overhead.
Best use is local, cost-conscious editing work: the accompanying paper, "Boogu-Image-0.1: Boosting Open Agentic Multimodal Generation via Understanding under a Minimal Budget" (arXiv 2607.13125) [6], explicitly targets minimal-budget generation. One caveat is early adoption — 1,119 downloads in the last 30 days as of 2026-09-25 [5], versus 323,535 for Qwen/Qwen-Image-Edit-2511 [1] — so expect fewer ready-made community workflows and quantized builds for now.
2. FireRedTeam/FireRed-Image-Edit-1.1
FireRedTeam/FireRed-Image-Edit-1.1 by FireRedTeam takes second place as the second-newest release in this field — published 2026-03-02, between Boogu/Boogu-Image-0.1-Edit (2026-06-16) and Qwen/Qwen-Image-Edit-2511 (2025-12-17) — because no candidate has a public benchmark yet [1][3][5]. Key specs: 20.4B parameters, 57.7 GB of BF16 weights, and an Apache-2.0 license [3], with a companion technical report on arXiv (2602.13344) [4].
Local runs on a 16 GB GPU require quantization: the 57.7 GB BF16 checkpoint [3] is far over budget, while a 4-bit load comes to about 10.2 GB estimated (20.4B parameters [3] x 0.5 bytes), which fits a 16 GB card with room for activations and the text/vision components.
Apache-2.0 licensing [3] makes FireRedTeam/FireRed-Image-Edit-1.1 a natural fit for teams that want a March-2026 release they can run and ship commercially on their own hardware. One caveat: no public benchmark yet scores this model, and adoption is early — 8,352 downloads in the last 30 days as of 2026-09-25, versus 323,535 for Qwen/Qwen-Image-Edit-2511 [1][3]. FireRedTeam/FireRed-Image-Edit-1.0 is superseded by this 1.1 release (2026-03-02, 20.4B) [3].
3. Qwen/Qwen-Image-Edit-2511
Qwen/Qwen-Image-Edit-2511, the editing release from the Qwen team first published 2025-12-17 [1], sits at number three purely on age: with no public benchmark yet scoring these models against each other, the ranking orders candidates by newest release first, then downloads, and 2025-12-17 [1] is the oldest release date among the three [1][3][5]. Its 323,535 downloads in the last 30 days [1] lead the field [1][3][5], but popularity only breaks ties here.
Specs: 20.4B parameters, BF16 weights at 57.7 GB, Apache-2.0 license [1], documented in the Qwen-Image Technical Report [2]. A 16 GB card cannot load the 57.7 GB BF16 checkpoint [1]; the practical route is a 4-bit quantization, estimated (params x 0.5 bytes) at about 10.2 GB for the weights from the 20.4B parameter count [1], leaving headroom on a 16 GB card for runtime overhead, on that estimate.
Best use: license-unrestricted image editing on one local card, since Apache-2.0 [1] permits commercial deployment and the estimated 4-bit weight budget (params x 0.5 bytes from 20.4B [1]) fits 16 GB hardware. One caveat: no public benchmark yet scores this generation, so third place reflects the newest-first ordering rather than any measured editing-quality gap behind FireRedTeam/FireRed-Image-Edit-1.1 or Boogu/Boogu-Image-0.1-Edit.
How much VRAM does a 20B image editing model need at 4-bit quantization?
A 20-billion-parameter image editing model needs roughly 10.2 GB of VRAM for its weights at 4-bit — estimated (params x 0.5 bytes) from the 20.4B parameter counts of Qwen/Qwen-Image-Edit-2511 [1] and FireRedTeam/FireRed-Image-Edit-1.1 [3] — so the weights fit on a 16 GB card with room to spare.
Full precision is not an option locally: both 20.4B models ship BF16 weights of 57.7 GB [1][3], far beyond a 16 GB GPU. Quantization is therefore the practical path: at 4-bit, the 20.4B-parameter models come to an estimated 10.2 GB (params × 0.5 bytes), leaving headroom for the rest of the pipeline on a 16 GB GPU [1][3].
Boogu/Boogu-Image-0.1-Edit gives more slack: at 10.3B parameters [5], weights come to about 5.15 GB estimated (params x 0.5 bytes), versus 38.5 GB at BF16 [5]. Boogu/Boogu-Image-0.1-Edit is also the newest release of the three, first published 2026-06-16 [5], ahead of FireRedTeam/FireRed-Image-Edit-1.1 (2026-03-02) [3] and Qwen/Qwen-Image-Edit-2511 (2025-12-17) [1].
Treat any quantized memory estimate as weight storage only. A realistic local build should budget memory headroom well beyond the bare weight estimate. No public benchmark yet scores any of these three candidates against each other, so fit and release recency — not measured quality — drive the comparison here.
Which open-weight image editing models run locally under an Apache-2.0 license?
Three open-weight image editing models run locally under an Apache-2.0 license: Boogu/Boogu-Image-0.1-Edit, FireRedTeam/FireRed-Image-Edit-1.1, and Qwen/Qwen-Image-Edit-2511 [1][3][5].
Ranking basis: no public benchmark scores any candidate, so the order below is newest release first, then downloads. Boogu/Boogu-Image-0.1-Edit ranks first as the newest release, published 2026-06-16, ahead of FireRed-Image-Edit-1.1 (2026-03-02) and Qwen-Image-Edit-2511 (2025-12-17) [1][3][5]; all three have no public benchmark yet.
Boogu/Boogu-Image-0.1-Edit carries 10.3B parameters [5], giving an estimated 4-bit memory of about 5.2 GB (params × 0.5 bytes); BF16 weights total 38.5 GB [5]. The estimate is the smallest of the three [1][3][5], suiting a 16 GB card. Downloads reached 1,119 in the last 30 days as of 2026-09-25 [5], and the paper "Boogu-Image-0.1: Boosting Open Agentic Multimodal Generation via Understanding under a Minimal Budget" (arXiv 2607.13125) documents the approach [6].
FireRedTeam/FireRed-Image-Edit-1.1 holds 20.4B parameters [3] — an estimated 4-bit memory of about 10.2 GB (params × 0.5 bytes) — with BF16 weights at 57.7 GB [3]. Downloads total 8,352 for the last 30 days as of 2026-09-25 [3]. The FireRed-Image-Edit-1.0 technical report (arXiv 2602.13344) documents the family [4], and FireRed-Image-Edit-1.0 itself is superseded by this release [3]. No public benchmark yet.
Qwen/Qwen-Image-Edit-2511, published 2025-12-17 [1], also runs 20.4B parameters [1] — an estimated 4-bit memory of about 10.2 GB (params × 0.5 bytes) — with BF16 weights at 57.7 GB [1]. Qwen-Image-Edit-2511 leads the group on downloads at 323,535 in the last 30 days as of 2026-09-25 [1][3][5], and the Qwen-Image Technical Report (arXiv 2508.02324) supplies implementation detail [2]. No public benchmark yet.
What is the newest open-weight image editing model release in 2026?
The newest open-weight image editing model release in 2026 is Boogu/Boogu-Image-0.1-Edit, first published on Hugging Face on 2026-06-16 [5]. Boogu-Image-0.1-Edit carries 10.3B parameters [5], BF16 weights at 38.5 GB [5], and an Apache-2.0 license [5]. At 4-bit, expect about 5.2 GB of VRAM, estimated (params × 0.5 bytes) from the 10.3B count [5] — the smallest footprint of the three candidates [5][3][1]. Last-30-day downloads: 1,119 [5]; no public benchmark yet [5].
FireRedTeam/FireRed-Image-Edit-1.1, first published on 2026-03-02, is the next-newest release [3]. The model has 20.4B parameters [3], BF16 weights at 57.7 GB [3], Apache-2.0 licensing [3], and 8,352 last-30-day downloads [3]; no public benchmark yet [3]. A 4-bit copy runs at roughly 10.2 GB, estimated (params × 0.5 bytes) from the 20.4B count [3]. FireRedTeam/FireRed-Image-Edit-1.0 is superseded by this 1.1 release [3].
Qwen/Qwen-Image-Edit-2511, first published on 2025-12-17, is the oldest release of the three [1][3][5]. The model matches FireRed-Image-Edit-1.1 at 20.4B parameters and 57.7 GB BF16 weights [1][3], keeps Apache-2.0 [1], and leads last-30-day downloads at 323,535 [1]; no public benchmark yet [1]. The Qwen-Image Technical Report (arXiv 2508.02324, 2025-08-04) documents the family [2].
Ranking note: no public benchmark scores any candidate, so the ordering is newest release first, then downloads — Boogu-Image-0.1-Edit (2026-06-16, 1,119 downloads) [5], FireRed-Image-Edit-1.1 (2026-03-02, 8,352) [3], Qwen-Image-Edit-2511 (2025-12-17, 323,535) [1]. Download counts serve only as a tiebreak, never as the reason a model leads.
Frequently Asked Questions
What is the best image editing model for a 16 GB GPU in 2026?
Boogu/Boogu-Image-0.1-Edit takes the top slot as the newest release of the group, first published on Hugging Face 2026-06-16 [5]. Ranking note: no public benchmark yet scores these candidates, so the ordering is newest release first, then downloads — Boogu-Image-0.1-Edit [5], FireRedTeam/FireRed-Image-Edit-1.1 (2026-03-02) [3], Qwen/Qwen-Image-Edit-2511 (2025-12-17) [1]. All three run under Apache-2.0 licenses [1][3][5].
Will a 20.4B image editing model actually fit on a 16 GB GPU?
Not at full precision. Qwen/Qwen-Image-Edit-2511 and FireRedTeam/FireRed-Image-Edit-1.1 each weigh 20.4B parameters with 57.7 GB of BF16 weights [1][3]. Quantized to 4-bit, 20.4B parameters [1][3] need roughly 10.2 GB estimated (params x 0.5 bytes), inside a 16 GB budget with headroom for runtime overhead. Boogu/Boogu-Image-0.1-Edit, at 10.3B parameters and 38.5 GB BF16 [5], is smaller still.
How much VRAM does Boogu-Image-0.1-Edit need locally?
Boogu/Boogu-Image-0.1-Edit has 10.3B parameters and 38.5 GB of BF16 weights [5]. At 4-bit, 10.3B parameters [5] come to about 5.2 GB estimated (params x 0.5 bytes), the smallest of the three on a 16 GB GPU — versus 20.4B parameters for Qwen/Qwen-Image-Edit-2511 and FireRedTeam/FireRed-Image-Edit-1.1 [1][3]. The license is Apache-2.0 [5].
Are these image editing models free to use commercially?
Yes. Qwen/Qwen-Image-Edit-2511, FireRedTeam/FireRed-Image-Edit-1.1, and Boogu/Boogu-Image-0.1-Edit all publish Apache-2.0 licenses on their Hugging Face model cards [1][3][5], permitting commercial use. FireRed-Image-Edit-1.1 (2026-03-02) [3] and Boogu-Image-0.1-Edit (2026-06-16) [5] are the current-generation releases of their families, each with a technical report on arXiv [4][6]. The Qwen-Image Technical Report (arXiv 2508.02324) documents that family [2].
Are there benchmark scores comparing Qwen-Image-Edit-2511, FireRed-Image-Edit-1.1, and Boogu-Image-0.1-Edit?
No public benchmark yet scores any of the three against the others. Each family has an arXiv technical report — Qwen-Image (2508.02324, 2025-08-04) [2], FireRed-Image-Edit-1.0 (2602.13344, 2026-02-12) [4], and Boogu-Image-0.1 (2607.13125, 2026-07-18) [6] — but none of these candidates carries a public leaderboard score, so no benchmark-based ordering is possible here.
Is FireRed-Image-Edit-1.0 still worth downloading?
Skip FireRedTeam/FireRed-Image-Edit-1.0, which is superseded by FireRedTeam/FireRed-Image-Edit-1.1 [3]. FireRedTeam/FireRed-Image-Edit-1.0 is superseded by FireRedTeam/FireRed-Image-Edit-1.1, released 2026-03-02 with 20.4B parameters and 57.7 GB BF16 weights [3]. At 4-bit, 20.4B parameters [3] comes to about 10.2 GB estimated (params x 0.5 bytes). FireRed-Image-Edit-1.1 logged 8,352 downloads in the 30 days to 2026-09-25 [3]; its technical report is arXiv 2602.13344 [4].
Sources
- Qwen/Qwen-Image-Edit-2511 model card (Hugging Face) — 2026-09-25
- Qwen-Image Technical Report — 2025-08-04
- FireRedTeam/FireRed-Image-Edit-1.1 model card (Hugging Face) — 2026-09-25
- FireRed-Image-Edit-1.0 Techinical Report — 2026-02-12
- Boogu/Boogu-Image-0.1-Edit model card (Hugging Face) — 2026-09-25
- Boogu-Image-0.1: Boosting Open Agentic Multimodal Generation via Understanding under a Minimal Budget — 2026-07-18