I think you've answered your own question: Historians use modern estimates to determine the death tolls in these conflicts and while these modern estimates do rely to an extent on contemporary Chinese accounts and records, they also have a considerably greater amount of work put into revising and adjusting their accuracy. Both of the examples you picked are actually also great examples for why historians really only want to look at modern estimates when speaking broadly and are not doing in-depth research to revise modern 'orthodox' figures.

Chinese civilization has an extremely long history when it comes to literary pursuits and part of that massive corpus of Chinese works include both governmental records and contemporary accounts written of historical events. Chinese records are also generally pretty good, as are historical accounts written by people of the time period. But during periods of extreme governmental upheaval, such as what we see in both the Ming-Qing transition (1644-c.1683) and the Taiping Civil War on the mid-19th century (it really was more than a mere 'rebellion') as well as other infamous examples of disorder such as the An Shi Rebellion (755-763), the accuracy and reliability of governmental records declines significantly. Part of this is due to the fact that the government often becomes ineffective in war-torn areas of China. The An Shi Rebellion is quite an extreme example of this as some estimates of the death toll based upon Tang dynasty census records place the death toll at upwards of 35 million people in the 8th century. But since the Tang's central authority immediately after the rebellion was crippled, it would probably be inaccurate to rely upon just governmental records alone to judge the true cost of the war. Same could be said of both the Ming-Qing Transition and the Taiping War. So when interpreting governmental records, historians need to also rely on data that comes from beyond what the written record delivers. This can include archaeological finds, contextual evidence, as well as other accounts not recorded in the official record.

Non-governmental records, while helpful, can also be inaccurate and need to also be approached with care and additional help when being addressed in modern analyses. For instance, the Ming-Qing War had a number of gruesome engagements such as the Siege of Yangzhou in which contemporary accounts, especially the famous Ten Days in Yangzhou, place the death toll exacted upon the city's defenders and population at over 800,000 but modern scholarship largely sees that estimate as an extreme exaggeration. The battle was definitely bloody - the Ten Days account note wells and ponds in the city dyed a deep red from the blood of the dead, babies trampled into the dirt by horses, whole army garrisons being killed to a man, etc., but modern scholarship does not simply rely on just eye-witness accounts to create casualty estimates.

Not the most in-depth of explanations but I hope this at least helps you contextualize what you're seeing.

Additional reading about the political nature of historical trauma in the formation of Modern China, which sort of relates to the accuracy of numbers you ask about, if interested.