Recently, we have deeply felt everyone’s expectations for DeepSeek in handling images, and the calls for it are quite high.
However, I must clarify that DeepSeek R1 is not a multi-modal large model, and its native support for images is indeed limited, unlike Doubao or ChatGPT, which can directly generate beautiful images.
But don’t be discouraged; as long as you master the methods, we can still make DeepSeek “play creatively” in image processing.
At the same time, using DeepSeek can help us generate more beautiful images, videos, even music, digital avatars, etc., achieving breakthroughs in different fields.
Let’s continue to look ahead.
01
Text to Image Generation
Two Steps to Easily Achieve
Text to image generation basically involves two steps:
1️⃣ First, propose a requirement for DeepSeek R1 to optimize the content and get the optimized text.
2️⃣ Then, propose a requirement for DeepSeek R1 to convert the optimized text into an image.
Among the several methods of generating images introduced below, the first step is the same; the difference lies in the second step.
In the first step, let DeepSeek R1 help us optimize the content.
Imitating the popular format from Xiaohongshu, the content is displayed as follows:
Title: DeepSeek Usage Tips
Tips:
1. Universal Question Template: Background + Requirement + Constraints (optional)
Example: My child is in the first year of junior high (background), how can I improve his English skills (requirement), no need to consider spoken language issues (constraints, optional)
2. Let DeepSeek “speak human language”
Example: I want to understand why DeepSeek’s cost is so low. Speak human language.
3. Imitate Character Responses
Example: Imitate Li Bai and write an acrostic poem “May All Go Well”.
The content optimized by DeepSeek for me is as follows:
How about it? Not bad, right?😁
Then we move on to the second step, converting the optimized text into an image. This step is relatively complex, but you won’t regret it after watching!
02
Revealing Various Image Generation Methods
■ 1. Generate SVG Images
The common image formats we see daily include png, jpg, webp, etc., while DeepSeek R1 currently only supports directly generating SVG format images.
Therefore, we can explicitly request DeepSeek to generate SVG images and put forward some specific requirements, such as color scheme, layout, etc.
For example, if I request “Convert the entire answer into an SVG image, requiring a minimalist color scheme,” DeepSeek outputs as follows:
Click Run HTML to pop up a page displaying the SVG image:
If the generated style meets your expectations, simply copy the entire block of content, create a new file, change the extension to svg, and paste the content.
If you feel the style is still not perfect, adjust it according to the color scheme, layout structure, etc., suggested by DeepSeek until you are satisfied.
■ 2. Generate PNG and JPG Images
If you particularly want to display images in png and jpg formats, there are ways to do that.
This type of requirement can generally be handled through specialized Python libraries for image processing or by writing an HTML file to generate images.
Don’t be intimidated by these technical terms; with DeepSeek’s help, we don’t need to write a single line of code.
For example, directly propose a requirement:
Display the entire answer in the style of a Xiaohongshu card, output in HTML, with the following requirements: 1. Each block of text corresponds to a card, each card provides a button to download as png, and the generated image does not include this button. 2. Do not adjust the text content. 3. Minimalist color scheme, aesthetically pleasing cards.
After thinking for a few seconds, DeepSeek directly returned the HTML code file to me.
Similarly, we just need to click the “Run HTML” button. For example, the style returned by DeepSeek is as follows.
Click “Save Image” to get the png image.
How about it? Isn’t it simple?
The images in the article have not been polished and are in their most basic style. If you are interested in producing more refined outputs and deeply cultivating in the self-media track, you can continue to look down. You can also join our DeepSeek free elite community to exchange and learn, receive the latest materials & the most comprehensive usage methods.
03
Expanding Ideas
Leveraging Third-Party Tools
In addition to the above methods, we can also use third-party AI tools for text-to-image conversion.
There are many free AI tools available on the market that can generate images based on prompts, and the quality of the prompts directly determines the quality of the images. Fortunately, DeepSeek is very good at generating prompts.
Midjourney
AI drawing section, the first recommended tool is still Midjourney, which not only has fast image generation speed but also produces top-level image quality. The latest image editor integrates Controlnet technology, but the freedom of control is relatively low, suitable for inspiration and material generation.
Flux
Flux is an AI image generation model launched by the Black Forest team, which generates images of very realistic quality, even comparable to Midjourney. Recently, a lot of beautiful images on Xiaohongshu were generated by the Flux model, and these images look like they were directly captured from the real world, with lifelike expressions and poses.
Recraft
Recraft is the most suitable AI image generation tool for designers, with strong text rendering capability. Although it does not support Chinese, it performs excellently in the design field. It is tailored for designers with four major functions: layout design, style transfer, vector image generation, and mockup generation, like a toolkit customized for designers to meet their creative needs in different scenarios.
In addition to the above methods, we can also leverage third-party AI text-to-image tools.
There are many free AI tools available on the market that can generate images based on prompts, and the quality of the prompts directly determines the quality of the images. Fortunately, DeepSeek is very good at generating prompts.
04
AI Video Tools
Keling AI
Keling AI is a video generation large model developed by the Kuaishou team, and it is currently the best-performing AI video generation tool on the market with the highest market share. The latest model 1.6 has significantly improved text responsiveness, better responding to descriptions of motion, timing actions, and camera movements, and the object movements are more reasonable, with more natural expressions. It acts like a professional film director, accurately understanding your ideas and transforming them into an excellent movie.
PixVerse
PixVerse video generation tool is developed by a domestic team, and the recently launched V3.5 model can generate a video in 10 seconds under Turbo mode, like an efficient Kuaishou, able to deliver works in a short time.
It has gained popularity both domestically and internationally mainly due to its “effect template” function, which allows you to generate effect videos by simply uploading images, selecting effects, and adding simple descriptions, such as the viral symbiote transformation effect in October.
Runway
Runway is considered the elder brother of video generation models, with a rich set of features, supporting video-to-image conversion, camera movements, expression control, etc. The generated video quality is high and can showcase complex scene changes and various film styles. It acts like an all-around film producer, able to meet professional creators’ needs in various aspects.
It has a particularly notable feature Act – One, which allows you to upload a video of a person to drive another character to replicate facial expressions 1:1, similar to motion capture in traditional films.
Above is the general content we compiled about DeepSeek’s handling of images and videos.
Of course, DeepSeek has more combinations that can bring qualitative benefits, which have been integrated into the workflows of countless elite professionals from prestigious schools and large companies.
This is just a glimpse in the vast galaxy; besides these, DeepSeek has many “game-changing” combinations.
Combined, whether you are a beginner, a novice, or an expert from various industries, you can ride the wind, either turning the tide or soaring high!
In this era of information explosion, AI brings a vast ocean of opportunities. Perhaps you are excited to harness the power of AI and make a mark in this era, but feel lost and unsure where to start amidst the overwhelming information?
Now, the opportunity has arrived! We sincerely invite you to join “How AI Thinks + Unlocking DeepSeek’s Game-Changing Combinations,” guiding you through the dense fog, avoiding the thorns and obstacles on the exploration path, with the least time cost, to seize this era’s “legendary secret”!
Now, a significant transformation is taking place in the field of large models, which can be regarded as a national-level development wave. We hope to lead you onto this train of the times, witnessing and participating in the industry’s transformation, ensuring that you do not miss this rare opportunity!
Top instructors from major companies will guide you in-depth exploration. Combining theory and practice, DeepSeek + multi-module tools’ game-changing combinations will help you achieve breakthroughs in multiple fields.
The first session of the course was fully booked! In anticipation, the second session of the elite version of the DeepSeek practical training camp sets sail.
This course is brought to you by two top instructors, featuring DeepSeek Practical Training Camp: How AI Thinks + Unlocking DeepSeek’s Game-Changing Combinations Theory + Practical Course.
We look forward to your joining us in steering destiny! Together, let’s witness this fascinating era.
Social Organizations
Industry Experience
Top Industry Instructors
Monetization Channels
Broad Development Platform
Practical Teaching
Introduction
BYD AI Head: Liu Huafang
South China University of Technology Alumni
Microsoft Certified Generative AI Engineer: Fang Yang Neo
One-on-one in-depth teaching.
Growth is not achieved overnight. The practical training camp has carefully crafted a progressive course system, from basic theory to cutting-edge applications, explained in a simple and profound way. Here, you can achieve step-by-step progress, mastering and applying this skill that will change the era, allowing your life to evolve with infinite possibilities.
We offer you far more than this. This will be a dazzling hall gathering elites from all walks of life, a high-value circle full of infinite possibilities. Offline courses include membership in the DeepSeek community, sharing courseware and learning materials; you will have opportunities for live interactive Q&A and practical training with DeepSeek’s large model tools, receiving one-on-one professional guidance from top instructors.
JOIN US
The Shenzhen Social Organization Exchange Service Display Point project aims to showcase the contributions made by social organizations in Shenzhen in serving society, promoting economic development, and fostering social harmony.
By selecting social organizations that are prominent in party building, highly representative, credible, and play a significant role in the construction of Shenzhen’s “Dual Zone,” we establish a platform for communication and display among social organizations across the city, showcasing their outstanding contributions to Shenzhen’s social and economic development and “Dual Zone” construction in various fields.
Through this initiative, we hope to encourage communication and cooperation between social organizations and promote their high-quality development.
| Federation Membership Department | 18211567364 |
| Federation Business Department | 18823495433 |
| Vocational Skills Training School | 13802218854 |
| Bay Area AI Computing Power Center | 18688993899 |
| VR Training Base | 13657272435 |
| Frontier Research Institute | 18188615433 |
| Low Altitude Economy Special Committee Training Base | 18823495433 |
Editor | Chen Zeyan
Responsible Editor | Liang Jinying
Proofreader | Tang Fei
Reviewer | Fu Mengjiao
Duty Editorial Committee | Liang Jinying